6.824 Lab 2: Raft 2B

Part 2B

We want Raft to keep a consistent, replicated log of operations. A call to Start() at the leader starts the process of adding a new operation to the log; the leader sends the new operation to the other servers in AppendEntries RPCs.我们希望Raft保持一致的、复制日志的操作。在leader处调用Start()将启动向日志添加新操作的过程;leader将新操作发送到附录条目rpc中的其他服务器。

Implement the leader and follower code to append new log entries. This will involve implementing Start(), completing the AppendEntries RPC structs, sending them, fleshing out the AppendEntry RPC handler, and advancing the commitIndex at the leader. Your first goal should be to pass the TestBasicAgree2B() test (in test_test.go). Once you have that working, you should get all the 2B tests to pass (go test -run 2B).

• You will need to implement the election restriction (section 5.4.1 in the paper).实现选举限制：如果日志不是新的，不给投票
• One way to fail the early Lab 2B tests is to hold un-needed elections, that is, an election even though the current leader is alive and can talk to all peers. This can prevent agreement in situations where the tester believes agreement is possible. Bugs in election timer management, or not sending out heartbeats immediately after winning an election, can cause un-needed elections.不必要的选举可能导致2b测试的过早失败
• You may need to write code that waits for certain events to occur. Do not write loops that execute continuously without pausing, since that will slow your implementation enough that it fails tests. You can wait efficiently with Go's channels, or Go's condition variables, or (if all else fails) by inserting a time.Sleep(10 * time.Millisecond) in each loop iteration.
• Give yourself time to rewrite your implementation in light of lessons learned about structuring concurrent code. In later labs you'll thank yourself for having Raft code that's as clear and clean as possible. For ideas, you can re-visit our structure, locking and guide, pages.

The tests for upcoming labs may fail your code if it runs too slowly. You can check how much real time and CPU time your solution uses with the time command. Here's some typical output for Lab 2B:

$ time go test -run 2B

Test (2B): basic agreement ...

... Passed --   0.5  5   28    3

Test (2B): agreement despite follower disconnection ...

... Passed --   3.9  3   69    7

Test (2B): no agreement if too many followers disconnect ...

... Passed --   3.5  5  144    4

Test (2B): concurrent Start()s ...

... Passed --   0.7  3   12    6

Test (2B): rejoin of partitioned leader ...

... Passed --   4.3  3  106    4

Test (2B): leader backs up quickly over incorrect follower logs ...

... Passed --  23.0  5 1302  102

Test (2B): RPC counts aren't too high ...

... Passed --   2.2  3   30   12

PASS

ok      raft    38.029s

real    0m38.511s

user    0m1.460s

sys     0m0.901s

$

The "ok raft 38.029s" means that Go measured the time taken for the 2B tests to be 38.029 seconds of real (wall-clock) time. The "user 0m1.460s" means that the code consumed 1.460 seconds of CPU time, or time spent actually executing instructions (rather than waiting or sleeping). If your solution uses much more than a minute of real time for the 2B tests, or much more than 5 seconds of CPU time, you may run into trouble later on. Look for time spent sleeping or waiting for RPC timeouts, loops that run without sleeping or waiting for conditions or channel messages, or large numbers of RPCs sent.

Be sure you pass the 2A and 2B tests before submitting Part 2B.

my homework code:

package raft

//
// this is an outline of the API that raft must expose to
// the service (or tester). see comments below for
// each of these functions for more details.
//
// rf = Make(...)
//   create a new Raft server.
// rf.Start(command interface{}) (index, term, isleader)
//   start agreement on a new log entry
// rf.GetState() (term, isLeader)
//   ask a Raft for its current term, and whether it thinks it is leader
// ApplyMsg
//   each time a new entry is committed to the log, each Raft peer
//   should send an ApplyMsg to the service (or tester)
//   in the same server.
//

import (
    "labrpc"
    "math/rand"
    "sync"
    "time"
)

// import "bytes"
// import "labgob"

//
// as each Raft peer becomes aware that successive log entries are
// committed, the peer should send an ApplyMsg to the service (or
// tester) on the same server, via the applyCh passed to Make(). set
// CommandValid to true to indicate that the ApplyMsg contains a newly
// committed log entry.
//
// in Lab 3 you'll want to send other kinds of messages (e.g.,
// snapshots) on the applyCh; at that point you can add fields to
// ApplyMsg, but set CommandValid to false for these other uses.
//
type ApplyMsg struct {
    CommandValid bool
    Command      interface{}
    CommandIndex int
}

type LogEntry struct {
    Command interface{}
    Term    int
}

const (
    Follower           int =
    Candidate          int =
    Leader             int =
    HEART_BEAT_TIMEOUT     =  //心跳超时，要求1秒10次，所以是100ms一次
)

//
// A Go object implementing a single Raft peer.
//
type Raft struct {
    mu        sync.Mutex          // Lock to protect shared access to this peer's state
    peers     []*labrpc.ClientEnd // RPC end points of all peers
    persister *Persister          // Object to hold this peer's persisted state
    me        int                 // this peer's index into peers[]

    // Your data here (2A, 2B, 2C).
    // Look at the paper's Figure 2 for a description of what
    // state a Raft server must maintain.
    electionTimer  *time.Timer   // 选举定时器
    heartbeatTimer *time.Timer   // 心跳定时器
    state          int           // 角色
    voteCount      int           //投票数
    applyCh        chan ApplyMsg // 提交通道

    currentTerm int        //latest term server has seen (initialized to 0 on first boot, increases monotonically)
    votedFor    int        //candidateId that received vote in current term (or null if none)
    log         []LogEntry //log entries; each entry contains command for state machine, and term when entry was received by leader (first index is 1)

    //Volatile state on all servers:
    commitIndex int //index of highest log entry known to be committed (initialized to 0, increases monotonically)
    lastApplied int //index of highest log entry applied to state machine (initialized to 0, increases monotonically)
    //Volatile state on leaders:(Reinitialized after election)
    nextIndex  []int //for each server, index of the next log entry to send to that server (initialized to leader last log index + 1)
    matchIndex []int //for each server, index of highest log entry known to be replicated on server (initialized to 0, increases monotonically)

}

// return currentTerm and whether this server
// believes it is the leader.
func (rf *Raft) GetState() (int, bool) {
    var term int
    var isleader bool
    // Your code here (2A).
    rf.mu.Lock()
    defer rf.mu.Unlock()
    term = rf.currentTerm
    isleader = rf.state == Leader
    return term, isleader
}

func (rf *Raft) persist() {
    // Your code here (2C).
    // Example:
    // w := new(bytes.Buffer)
    // e := labgob.NewEncoder(w)
    // e.Encode(rf.xxx)
    // e.Encode(rf.yyy)
    // data := w.Bytes()
    // rf.persister.SaveRaftState(data)
}

//
// restore previously persisted state.
//
func (rf *Raft) readPersist(data []byte) {
    if data == nil || len(data) <  { // bootstrap without any state?
        return
    }
    // Your code here (2C).
    // Example:
    // r := bytes.NewBuffer(data)
    // d := labgob.NewDecoder(r)
    // var xxx
    // var yyy
    // if d.Decode(&xxx) != nil ||
    //    d.Decode(&yyy) != nil {
    //   error...
    // } else {
    //   rf.xxx = xxx
    //   rf.yyy = yyy
    // }
}

//
// example RequestVote RPC arguments structure.
// field names must start with capital letters!
//
type RequestVoteArgs struct {
    // Your data here (2A, 2B).
    Term         int //candidate’s term
    CandidateId  int //candidate requesting vote
    LastLogIndex int //index of candidate’s last log entry (§5.4)
    LastLogTerm  int //term of candidate’s last log entry (§5.4)
}

//
// example RequestVote RPC reply structure.
// field names must start with capital letters!
//
type RequestVoteReply struct {
    // Your data here (2A).
    Term        int  //currentTerm, for candidate to update itself
    VoteGranted bool //true means candidate received vote
}

//
// example RequestVote RPC handler.
//
func (rf *Raft) RequestVote(args *RequestVoteArgs, reply *RequestVoteReply) {
    // Your code here (2A, 2B).
    rf.mu.Lock()
    defer rf.mu.Unlock()
    DPrintf("Candidate[raft%v][term:%v] request vote: raft%v[%v] 's term%v\n", args.CandidateId, args.Term, rf.me, rf.state, rf.currentTerm)
    if args.Term < rf.currentTerm ||
        (args.Term == rf.currentTerm && rf.votedFor != - && rf.votedFor != args.CandidateId) {
        reply.Term = rf.currentTerm
        reply.VoteGranted = false
        return
    }

    if args.Term > rf.currentTerm {
        rf.currentTerm = args.Term
        rf.switchStateTo(Follower)
    }

    // 2B: candidate's vote should be at least up-to-date as receiver's log
    // "up-to-date" is defined in thesis 5.4.1
    lastLogIndex := len(rf.log) -
    if args.LastLogTerm < rf.log[lastLogIndex].Term ||
        (args.LastLogTerm == rf.log[lastLogIndex].Term &&
            args.LastLogIndex < (lastLogIndex)) {
        // Receiver is more up-to-date, does not grant vote
        reply.Term = rf.currentTerm
        reply.VoteGranted = false
        return
    }

    rf.votedFor = args.CandidateId
    reply.Term = rf.currentTerm
    reply.VoteGranted = true
    // reset timer after grant vote
    rf.electionTimer.Reset(randTimeDuration())
}

type AppendEntriesArgs struct {
    Term         int        //leader’s term
    LeaderId     int        //so follower can redirect clients
    PrevLogIndex int        //index of log entry immediately preceding new ones
    PrevLogTerm  int        //term of prevLogIndex entry
    Entries      []LogEntry //log entries to store (empty for heartbeat; may send more than one for efficiency)
    LeaderCommit int        //leader’s commitIndex
}

type AppendEntriesReply struct {
    Term    int  //currentTerm, for leader to update itself
    Success bool //true if follower contained entry matching prevLogIndex and prevLogTerm
}

func (rf *Raft) AppendEntries(args *AppendEntriesArgs, reply *AppendEntriesReply) {
    rf.mu.Lock()
    defer rf.mu.Unlock()
    DPrintf("leader[raft%v][term:%v] beat term:%v [raft%v][%v]\n", args.LeaderId, args.Term, rf.currentTerm, rf.me, rf.state)
    reply.Success = true

    // 1. Reply false if term < currentTerm (§5.1)
    if args.Term < rf.currentTerm {
        reply.Success = false
        reply.Term = rf.currentTerm
        return
    }
    //If RPC request or response contains term T > currentTerm:set currentTerm = T, convert to follower (§5.1)
    if args.Term > rf.currentTerm {
        rf.currentTerm = args.Term
        rf.switchStateTo(Follower)
    }

    // reset election timer even log does not match
    // args.LeaderId is the current term's Leader
    rf.electionTimer.Reset(randTimeDuration())

    // 2. Reply false if log doesn’t contain an entry at prevLogIndex
    // whose term matches prevLogTerm (§5.3)
    lastLogIndex := len(rf.log) -
    if lastLogIndex < args.PrevLogIndex {
        reply.Success = false
        reply.Term = rf.currentTerm
        return
    }

    // 3. If an existing entry conflicts with a new one (same index
    // but different terms), delete the existing entry and all that
    // follow it (§5.3)
    if rf.log[(args.PrevLogIndex)].Term != args.PrevLogTerm {
        reply.Success = false
        reply.Term = rf.currentTerm
        return
    }

    // 4. Append any new entries not already in the log
    // compare from rf.log[args.PrevLogIndex + 1]
    unmatch_idx := -
    for idx := range args.Entries {
        if len(rf.log) < (args.PrevLogIndex++idx) ||
            rf.log[(args.PrevLogIndex++idx)].Term != args.Entries[idx].Term {
            // unmatch log found
            unmatch_idx = idx
            break
        }
    }

    if unmatch_idx != - {
        // there are unmatch entries
        // truncate unmatch Follower entries, and apply Leader entries
        rf.log = rf.log[:(args.PrevLogIndex +  + unmatch_idx)]
        rf.log = append(rf.log, args.Entries[unmatch_idx:]...)
    }

    //5. If leaderCommit > commitIndex, set commitIndex = min(leaderCommit, index of last new entry)
    if args.LeaderCommit > rf.commitIndex {
        rf.setCommitIndex(min(args.LeaderCommit, len(rf.log)-))
    }

    reply.Success = true
}

//
// example code to send a RequestVote RPC to a server.
// server is the index of the target server in rf.peers[].
// expects RPC arguments in args.
// fills in *reply with RPC reply, so caller should
// pass &reply.
// the types of the args and reply passed to Call() must be
// the same as the types of the arguments declared in the
// handler function (including whether they are pointers).
//
// The labrpc package simulates a lossy network, in which servers
// may be unreachable, and in which requests and replies may be lost.
// Call() sends a request and waits for a reply. If a reply arrives
// within a timeout interval, Call() returns true; otherwise
// Call() returns false. Thus Call() may not return for a while.
// A false return can be caused by a dead server, a live server that
// can't be reached, a lost request, or a lost reply.
//
// Call() is guaranteed to return (perhaps after a delay) *except* if the
// handler function on the server side does not return.  Thus there
// is no need to implement your own timeouts around Call().
//
// look at the comments in ../labrpc/labrpc.go for more details.
//
// if you're having trouble getting RPC to work, check that you've
// capitalized all field names in structs passed over RPC, and
// that the caller passes the address of the reply struct with &, not
// the struct itself.
//
func (rf *Raft) sendRequestVote(server int, args *RequestVoteArgs, reply *RequestVoteReply) bool {
    ok := rf.peers[server].Call("Raft.RequestVote", args, reply)
    return ok
}

func (rf *Raft) sendAppendEntries(server int, args *AppendEntriesArgs, reply *AppendEntriesReply) bool {
    ok := rf.peers[server].Call("Raft.AppendEntries", args, reply)
    return ok
}

//
// the service using Raft (e.g. a k/v server) wants to start
// agreement on the next command to be appended to Raft's log. if this
// server isn't the leader, returns false. otherwise start the
// agreement and return immediately. there is no guarantee that this
// command will ever be committed to the Raft log, since the leader
// may fail or lose an election. even if the Raft instance has been killed,
// this function should return gracefully.
//
// the first return value is the index that the command will appear at
// if it's ever committed. the second return value is the current
// term. the third return value is true if this server believes it is
// the leader.
//
func (rf *Raft) Start(command interface{}) (int, int, bool) {
    index := -
    term := -
    isLeader := true

    // Your code here (2B).
    rf.mu.Lock()
    defer rf.mu.Unlock()
    term = rf.currentTerm
    isLeader = rf.state == Leader
    if isLeader {
        rf.log = append(rf.log, LogEntry{Command: command, Term: term})
        index = len(rf.log) -
        rf.matchIndex[rf.me] = index
        rf.nextIndex[rf.me] = index +
    }

    return index, term, isLeader
}

//
// the tester calls Kill() when a Raft instance won't
// be needed again. you are not required to do anything
// in Kill(), but it might be convenient to (for example)
// turn off debug output from this instance.
//
func (rf *Raft) Kill() {
    // Your code here, if desired.
}

//
// the service or tester wants to create a Raft server. the ports
// of all the Raft servers (including this one) are in peers[]. this
// server's port is peers[me]. all the servers' peers[] arrays
// have the same order. persister is a place for this server to
// save its persistent state, and also initially holds the most
// recent saved state, if any. applyCh is a channel on which the
// tester or service expects Raft to send ApplyMsg messages.
// Make() must return quickly, so it should start goroutines
// for any long-running work.
//
func Make(peers []*labrpc.ClientEnd, me int,
    persister *Persister, applyCh chan ApplyMsg) *Raft {
    rf := &Raft{}
    rf.peers = peers
    rf.persister = persister
    rf.me = me

    // Your initialization code here (2A, 2B, 2C).
    rf.state = Follower
    rf.votedFor = -
    rf.heartbeatTimer = time.NewTimer(HEART_BEAT_TIMEOUT * time.Millisecond)
    rf.electionTimer = time.NewTimer(randTimeDuration())

    rf.applyCh = applyCh
    rf.log = make([]LogEntry, ) // start from index 1

    rf.nextIndex = make([]int, len(rf.peers))
    rf.matchIndex = make([]int, len(rf.peers))

    //以定时器的维度重写background逻辑
    go func() {
        for {
            select {
            case <-rf.electionTimer.C:
                rf.mu.Lock()
                switch rf.state {
                case Follower:
                    rf.switchStateTo(Candidate)
                case Candidate:
                    rf.startElection()
                }
                rf.mu.Unlock()

            case <-rf.heartbeatTimer.C:
                rf.mu.Lock()
                if rf.state == Leader {
                    rf.heartbeats()
                    rf.heartbeatTimer.Reset(HEART_BEAT_TIMEOUT * time.Millisecond)
                }
                rf.mu.Unlock()
            }
        }
    }()

    // initialize from state persisted before a crash
    rf.readPersist(persister.ReadRaftState())

    return rf
}

func randTimeDuration() time.Duration {
    return time.Duration(HEART_BEAT_TIMEOUT*+rand.Intn(HEART_BEAT_TIMEOUT)) * time.Millisecond
}

//切换状态，调用者需要加锁
func (rf *Raft) switchStateTo(state int) {
    if state == rf.state {
        return
    }
    DPrintf("Term %d: server %d convert from %v to %v\n", rf.currentTerm, rf.me, rf.state, state)
    rf.state = state
    switch state {
    case Follower:
        rf.heartbeatTimer.Stop()
        rf.electionTimer.Reset(randTimeDuration())
        rf.votedFor = -

    case Candidate:
        //成为候选人后立马进行选举
        rf.startElection()

    case Leader:
        // initialized to leader last log index + 1
        for i := range rf.nextIndex {
            rf.nextIndex[i] = (len(rf.log))
        }
        for i := range rf.matchIndex {
            rf.matchIndex[i] =
        }

        rf.electionTimer.Stop()
        rf.heartbeats()
        rf.heartbeatTimer.Reset(HEART_BEAT_TIMEOUT * time.Millisecond)
    }
}

// 发送心跳包，调用者需要加锁
func (rf *Raft) heartbeats() {
    for i := range rf.peers {
        if i != rf.me {
            go rf.heartbeat(i)
        }
    }
}

func (rf *Raft) heartbeat(server int) {
    rf.mu.Lock()
    if rf.state != Leader {
        rf.mu.Unlock()
        return
    }

    prevLogIndex := rf.nextIndex[server] - 

    // use deep copy to avoid race condition
    // when override log in AppendEntries()
    entries := make([]LogEntry, len(rf.log[(prevLogIndex+):]))
    copy(entries, rf.log[(prevLogIndex+):])

    args := AppendEntriesArgs{
        Term:         rf.currentTerm,
        LeaderId:     rf.me,
        PrevLogIndex: prevLogIndex,
        PrevLogTerm:  rf.log[(prevLogIndex)].Term,
        Entries:      entries,
        LeaderCommit: rf.commitIndex,
    }
    rf.mu.Unlock()

    var reply AppendEntriesReply
    if rf.sendAppendEntries(server, &args, &reply) {
        rf.mu.Lock()
        if rf.state != Leader {
            rf.mu.Unlock()
            return
        }
        // If last log index ≥ nextIndex for a follower: send
        // AppendEntries RPC with log entries starting at nextIndex
        // • If successful: update nextIndex and matchIndex for
        // follower (§5.3)
        // • If AppendEntries fails because of log inconsistency:
        // decrement nextIndex and retry (§5.3)
        if reply.Success {
            // successfully replicated args.Entries
            rf.matchIndex[server] = args.PrevLogIndex + len(args.Entries)
            rf.nextIndex[server] = rf.matchIndex[server] + 

            // If there exists an N such that N > commitIndex, a majority
            // of matchIndex[i] ≥ N, and log[N].term == currentTerm:
            // set commitIndex = N (§5.3, §5.4).
            for N := (len(rf.log) - ); N > rf.commitIndex; N-- {
                count :=
                for _, matchIndex := range rf.matchIndex {
                    if matchIndex >= N {
                        count +=
                    }
                }

                if count > len(rf.peers)/ {
                    // most of nodes agreed on rf.log[i]
                    rf.setCommitIndex(N)
                    break
                }
            }

        } else {
            if reply.Term > rf.currentTerm {
                rf.currentTerm = reply.Term
                rf.switchStateTo(Follower)
            } else {
                //如果走到这个分支，那一定是需要前推
                rf.nextIndex[server] = args.PrevLogIndex -
            }
        }
        rf.mu.Unlock()
    }
}

// 开始选举，调用者需要加锁
func (rf *Raft) startElection() {

    //    DPrintf("raft%v is starting election\n", rf.me)
    rf.currentTerm +=
    rf.votedFor = rf.me //vote for me
    rf.voteCount =
    rf.electionTimer.Reset(randTimeDuration())

    for i := range rf.peers {
        if i != rf.me {
            go func(peer int) {
                rf.mu.Lock()
                lastLogIndex := len(rf.log) -
                args := RequestVoteArgs{
                    Term:         rf.currentTerm,
                    CandidateId:  rf.me,
                    LastLogIndex: lastLogIndex,
                    LastLogTerm:  rf.log[lastLogIndex].Term,
                }
                // DPrintf("raft%v[%v] is sending RequestVote RPC to raft%v\n", rf.me, rf.state, peer)
                rf.mu.Unlock()
                var reply RequestVoteReply
                if rf.sendRequestVote(peer, &args, &reply) {
                    rf.mu.Lock()
                    defer rf.mu.Unlock()
                    if reply.Term > rf.currentTerm {
                        rf.currentTerm = reply.Term
                        rf.switchStateTo(Follower)
                    }
                    if reply.VoteGranted && rf.state == Candidate {
                        rf.voteCount++
                        if rf.voteCount > len(rf.peers)/ {
                            rf.switchStateTo(Leader)
                        }
                    }
                }
            }(i)
        }
    }
}

//
// several setters, should be called with a lock
//
func (rf *Raft) setCommitIndex(commitIndex int) {
    rf.commitIndex = commitIndex
    // apply all entries between lastApplied and committed
    // should be called after commitIndex updated
    if rf.commitIndex > rf.lastApplied {
        DPrintf("%v apply from index %d to %d", rf, rf.lastApplied+, rf.commitIndex)
        entriesToApply := append([]LogEntry{}, rf.log[(rf.lastApplied+):(rf.commitIndex+)]...)

        go func(startIdx int, entries []LogEntry) {
            for idx, entry := range entries {
                var msg ApplyMsg
                msg.CommandValid = true
                msg.Command = entry.Command
                msg.CommandIndex = startIdx + idx
                rf.applyCh <- msg
                // do not forget to update lastApplied index
                // this is another goroutine, so protect it with lock
                rf.mu.Lock()
                if rf.lastApplied < msg.CommandIndex {
                    rf.lastApplied = msg.CommandIndex
                }
                rf.mu.Unlock()
            }
        }(rf.lastApplied+, entriesToApply)
    }
}
func min(x, y int) int {
    if x < y {
        return x
    } else {
        return y
    }
}

test

test -race