6.824 Lab 3: Fault-tolerant Key/Value Service 3B

Part B: Key/value service with log compaction

Do a git pull to get the latest lab software.

As things stand now with your lab code, a rebooting server replays the complete Raft log in order to restore its state. However, it's not practical for a long-running server to remember the complete Raft log forever. Instead, you'll modify Raft and kvserver to cooperate to save space: from time to time kvserver will persistently store a "snapshot" of its current state, and Raft will discard log entries that precede the snapshot. When a server restarts (or falls far behind the leader and must catch up), the server first installs a snapshot and then replays log entries from after the point at which the snapshot was created. Section 7 of the extended Raft paper outlines the scheme; you will have to design the details.根据您的实验代码，重新启动的服务器将重播完整的筏日志以恢复其状态。但是，对于长时间运行的服务器来说，永远记住完整的筏日志是不现实的。相反，您将修改Raft和kvserver以节省空间:kvserver将持续地存储当前状态的“快照”，而Raft将丢弃快照之前的日志条目。当服务器重新启动(或远远落后于领先者，必须迎头赶上)时，服务器首先安装快照，然后从创建快照的位置开始重播日志条目。You should spend some time figuring out what the interface will be between your Raft library and your service so that your Raft library can discard log entries. Think about how your Raft will operate while storing only the tail of the log, and how it will discard old log entries. You should discard them in a way that allows the Go garbage collector to free and re-use the memory; this requires that there be no reachable references (pointers) to the discarded log entries.您应该花一些时间来确定您的Raft库和服务之间的接口，以便您的Raft库可以丢弃日志条目。考虑一下您的Raft将如何操作，而只存储日志的尾部，以及它将如何丢弃旧的日志条目。您应该以允许Go垃圾收集器释放和重用内存的方式丢弃它们;这要求对丢弃的日志条目没有可到达的引用(指针)。

The tester passes maxraftstate to your StartKVServer(). maxraftstate indicates the maximum allowed size of your persistent Raft state in bytes (including the log, but not including snapshots). You should compare maxraftstate to persister.RaftStateSize(). Whenever your key/value server detects that the Raft state size is approaching this threshold, it should save a snapshot, and tell the Raft library that it has snapshotted, so that Raft can discard old log entries. If maxraftstate is -1, you do not have to snapshot.当您的键/值服务器检测到Raft的状态大小接近这个阈值时，它应该保存一个快照，并告诉筏库它已经快照了，以便Raft可以丢弃旧的日志条目。如果maxraftstate为-1，则不必快照。

Your raft.go probably keeps the entire log in a Go slice. Modify it so that it can be given a log index, discard the entries before that index, and continue operating while storing only log entries after that index. Make sure you pass all the Raft tests after making these changes.

Modify your kvserver so that it detects when the persisted Raft state grows too large, and then hands a snapshot to Raft and tells Raft that it can discard old log entries. Raft should save each snapshot with persister.SaveStateAndSnapshot() (don't use files). A kvserver instance should restore the snapshot from the persister when it re-starts.kvserver实例应该在重新启动时从persister恢复快照。

• You can test your Raft and kvserver's ability to operate with a trimmed log, and its ability to re-start from the combination of a kvserver snapshot and persisted Raft state, by running the Lab 3A tests while artificially setting maxraftstate to 1.
• Think about when a kvserver should snapshot its state and what should be included in the snapshot. Raft must store each snapshot in the persister object using SaveStateAndSnapshot(), along with corresponding Raft state. You can read the latest stored snapshot using ReadSnapshot().
• Your kvserver must be able to detect duplicated operations in the log across checkpoints, so any state you are using to detect them must be included in the snapshots. Remember to capitalize all fields of structures stored in the snapshot.
• You are allowed to add methods to your Raft so that kvserver can manage the process of trimming the Raft log and manage kvserver snapshots.

Modify your Raft leader code to send an InstallSnapshot RPC to a follower when the leader has discarded the log entries the follower needs. When a follower receives an InstallSnapshot RPC, your Raft code will need to send the included snapshot to its kvserver. You can use the applyCh for this purpose, by adding new fields to ApplyMsg. Your solution is complete when it passes all of the Lab 3 tests.

The maxraftstate limit applies to the GOB-encoded bytes your Raft passes to persister.SaveRaftState().

• You should send the entire snapshot in a single InstallSnapshot RPC. You do not have to implement Figure 13's offset mechanism for splitting up the snapshot.
• Make sure you pass TestSnapshotRPC before moving on to the other Snapshot tests.
• A reasonable amount of time to take for the Lab 3 tests is 400 seconds of real time and 700 seconds of CPU time. Further, go test -run TestSnapshotSize should take less than 20 seconds of real time.

Your code should pass the 3B tests (as in the example here) as well as the 3A tests.

$ go test -run 3B

Test: InstallSnapshot RPC (3B) ...

  ... Passed --   1.5  3   163   63

Test: snapshot size is reasonable (3B) ...

  ... Passed --   0.4  3  2407  800

Test: restarts, snapshots, one client (3B) ...

  ... Passed --  19.2  5 123372 24718

Test: restarts, snapshots, many clients (3B) ...

  ... Passed --  18.9  5 127387 58305

Test: unreliable net, snapshots, many clients (3B) ...

  ... Passed --  16.3  5  4485 1053

Test: unreliable net, restarts, snapshots, many clients (3B) ...

  ... Passed --  20.7  5  4802 1005

Test: unreliable net, restarts, partitions, snapshots, many clients (3B) ...

  ... Passed --  27.1  5  3281  535

Test: unreliable net, restarts, partitions, snapshots, many clients, linearizability checks (3B) ...

  ... Passed --  25.0  7 11344  748

PASS

ok      kvraft  129.114s

my homework code:raft.go

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

    // to send kv snapshot to kv server
    CommandData []byte // 3B
}

type LogEntry struct {
    Command interface{}
    Term    int
}

const (
    Follower           int = 1
    Candidate          int = 2
    Leader             int = 3
    HEART_BEAT_TIMEOUT     = 100 //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

    snapshottedIndex int // 3B 归档位置

    //Persistent state on all servers:
    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) encodeRaftState() []byte {
    w := new(bytes.Buffer)
    e := labgob.NewEncoder(w)
    e.Encode(rf.currentTerm)
    e.Encode(rf.votedFor)
    e.Encode(rf.log)
    e.Encode(rf.snapshottedIndex)
    return w.Bytes()
}

func (rf *Raft) persist() {
    // Your code here (2C).
    // Example:
    // w := new(bytes.Buffer)
    // e := labgob.NewEncoder(w)
    // e.Encode(rf.currentTerm)
    // e.Encode(rf.votedFor)
    // e.Encode(rf.log)
    // e.Encode(rf.snapshottedIndex)
    // data := w.Bytes()
    rf.persister.SaveRaftState(rf.encodeRaftState())
}
func (rf *Raft) GetRaftStateSize() int {
    return rf.persister.RaftStateSize()
}

//
// restore previously persisted state.
//
func (rf *Raft) readPersist(data []byte) {
    if data == nil || len(data) < 1 { // bootstrap without any state?
        return
    }
    // Your code here (2C).
    // Example:
    r := bytes.NewBuffer(data)
    d := labgob.NewDecoder(r)
    var currentTerm int
    var votedFor int
    var log []LogEntry
    var snapshottedIndex int
    if d.Decode(&currentTerm) != nil ||
        d.Decode(&votedFor) != nil ||
        d.Decode(&log) != nil ||
        d.Decode(&snapshottedIndex) != nil {
        // error...
        panic("fail to decode state")
    } else {
        rf.currentTerm = currentTerm
        rf.votedFor = votedFor
        rf.log = log
        rf.snapshottedIndex = snapshottedIndex
        // for lab 3b, we need to set them at the first index
        // i.e., 0 if snapshot is disabled
        rf.commitIndex = snapshottedIndex
        rf.lastApplied = snapshottedIndex
    }
}

//
// 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()
    defer rf.persist() // 改动需要持久化
    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 != -1 && 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) - 1
    if args.LastLogTerm < rf.log[lastLogIndex].Term ||
        (args.LastLogTerm == rf.log[lastLogIndex].Term &&
            args.LastLogIndex < rf.getAbsoluteLogIndex(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

    //Figure 8: A time sequence showing why a leader cannot determine commitment using log entries from older terms. In
    // (a) S1 is leader and partially replicates the log entry at index
    // 2. In (b) S1 crashes; S5 is elected leader for term 3 with votes
    // from S3, S4, and itself, and accepts a different entry at log
    // index 2. In (c) S5 crashes; S1 restarts, is elected leader, and
    // continues replication. At this point, the log entry from term 2
    // has been replicated on a majority of the servers, but it is not
    // committed. If S1 crashes as in (d), S5 could be elected leader
    // (with votes from S2, S3, and S4) and overwrite the entry with
    // its own entry from term 3. However, if S1 replicates an entry from its current term on a majority of the servers before
    // crashing, as in (e), then this entry is committed (S5 cannot
    // win an election). At this point all preceding entries in the log
    // are committed as well.
    ConflictTerm  int // 2C
    ConflictIndex int // 2C
}

func (rf *Raft) AppendEntries(args *AppendEntriesArgs, reply *AppendEntriesReply) {
    rf.mu.Lock()
    defer rf.mu.Unlock()
    defer rf.persist() // 改动需要持久化
    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())

    if args.PrevLogIndex <= rf.snapshottedIndex {
        reply.Success = true

        // sync log if needed
        if args.PrevLogIndex+len(args.Entries) > rf.snapshottedIndex {
            // if snapshottedIndex == prevLogIndex, all log entries should be added.
            startIdx := rf.snapshottedIndex - args.PrevLogIndex
            // only keep the last snapshotted one
            rf.log = rf.log[:1]
            rf.log = append(rf.log, args.Entries[startIdx:]...)
        }

        return
    }

    // 2. Reply false if log doesn’t contain an entry at prevLogIndex
    // whose term matches prevLogTerm (§5.3)
    lastLogIndex := rf.getAbsoluteLogIndex(len(rf.log) - 1)
    if lastLogIndex < args.PrevLogIndex {
        reply.Success = false
        reply.Term = rf.currentTerm
        // optimistically thinks receiver's log matches with Leader's as a subset
        reply.ConflictIndex = len(rf.log)
        // no conflict term
        reply.ConflictTerm = -1
        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[rf.getRelativeLogIndex(args.PrevLogIndex)].Term != args.PrevLogTerm {
        reply.Success = false
        reply.Term = rf.currentTerm
        // receiver's log in certain term unmatches Leader's log
        reply.ConflictTerm = rf.log[rf.getRelativeLogIndex(args.PrevLogIndex)].Term

        // expecting Leader to check the former term
        // so set ConflictIndex to the first one of entries in ConflictTerm
        conflictIndex := args.PrevLogIndex
        // apparently, since rf.log[0] are ensured to match among all servers
        // ConflictIndex must be > 0, safe to minus 1
        for rf.log[rf.getRelativeLogIndex(conflictIndex-1)].Term == reply.ConflictTerm {
            conflictIndex--
            if conflictIndex == rf.snapshottedIndex+1 {
                // this may happen after snapshot,
                // because the term of the first log may be the current term
                // before lab 3b this is not going to happen, since rf.log[0].Term = 0
                break
            }
        }
        reply.ConflictIndex = conflictIndex
        return
    }

    // 4. Append any new entries not already in the log
    // compare from rf.log[args.PrevLogIndex + 1]
    unmatch_idx := -1
    for idx := range args.Entries {
        if len(rf.log) < rf.getRelativeLogIndex(args.PrevLogIndex+2+idx) ||
            rf.log[rf.getRelativeLogIndex(args.PrevLogIndex+1+idx)].Term != args.Entries[idx].Term {
            // unmatch log found
            unmatch_idx = idx
            break
        }
    }

    if unmatch_idx != -1 {
        // there are unmatch entries
        // truncate unmatch Follower entries, and apply Leader entries
        rf.log = rf.log[:rf.getRelativeLogIndex(args.PrevLogIndex+1+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, rf.getAbsoluteLogIndex(len(rf.log)-1)))
    }

    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 := -1
    term := -1
    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})
        rf.persist() // 改动需要持久化
        index = rf.getAbsoluteLogIndex(len(rf.log) - 1)
        rf.matchIndex[rf.me] = index
        rf.nextIndex[rf.me] = index + 1
    }

    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 = -1
    rf.heartbeatTimer = time.NewTimer(HEART_BEAT_TIMEOUT * time.Millisecond)
    rf.electionTimer = time.NewTimer(randTimeDuration())

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

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

    rf.nextIndex = make([]int, len(rf.peers))
    //for persist
    for i := range rf.nextIndex {
        // initialized to leader last log index + 1
        rf.nextIndex[i] = len(rf.log)
    }
    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()
            }
        }
    }()

    return rf
}

func randTimeDuration() time.Duration {
    return time.Duration(HEART_BEAT_TIMEOUT*3+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 = -1
    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] = rf.getAbsoluteLogIndex(len(rf.log))
        }
        for i := range rf.matchIndex {
            rf.matchIndex[i] = rf.snapshottedIndex //3B
        }

        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] - 1

    if prevLogIndex < rf.snapshottedIndex {
        // leader has discarded log entries the follower needs
        // send snapshot to follower and retry later
        rf.mu.Unlock()
        rf.syncSnapshotWith(server)
        return
    }

    // use deep copy to avoid race condition
    // when override log in AppendEntries()
    entries := make([]LogEntry, len(rf.log[rf.getRelativeLogIndex(prevLogIndex+1):]))
    copy(entries, rf.log[rf.getRelativeLogIndex(prevLogIndex+1):])

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

    var reply AppendEntriesReply
    if rf.sendAppendEntries(server, &args, &reply) {
        rf.mu.Lock()
        defer rf.mu.Unlock()
        if rf.state != Leader {
            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] + 1

            // 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 := rf.getAbsoluteLogIndex(len(rf.log) - 1); N > rf.commitIndex; N-- {
                count := 0
                for _, matchIndex := range rf.matchIndex {
                    if matchIndex >= N {
                        count += 1
                    }
                }

                if count > len(rf.peers)/2 {
                    // 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)
                rf.persist() // 改动需要持久化
            } else {
                //如果走到这个分支，那一定是需要前推(优化前推)
                rf.nextIndex[server] = reply.ConflictIndex

                // if term found, override it to
                // the first entry after entries in ConflictTerm
                if reply.ConflictTerm != -1 {
                    for i := args.PrevLogIndex; i >= rf.snapshottedIndex+1; i-- {
                        if rf.log[rf.getRelativeLogIndex(i-1)].Term == reply.ConflictTerm {
                            // in next trial, check if log entries in ConflictTerm matches
                            rf.nextIndex[server] = i
                            break
                        }
                    }
                }
                //和等待下一轮执行相比，直接retry并没有明显优势，
                // go rf.heartbeat(server)
            }
        }
        // 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 += 1
    rf.votedFor = rf.me //vote for me
    rf.persist()        // 改动需要持久化
    rf.voteCount = 1
    rf.electionTimer.Reset(randTimeDuration())

    for i := range rf.peers {
        if i != rf.me {
            go func(peer int) {
                rf.mu.Lock()
                lastLogIndex := len(rf.log) - 1
                args := RequestVoteArgs{
                    Term:         rf.currentTerm,
                    CandidateId:  rf.me,
                    LastLogIndex: rf.getAbsoluteLogIndex(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)
                        rf.persist() // 改动需要持久化
                    }
                    if reply.VoteGranted && rf.state == Candidate {
                        rf.voteCount++
                        if rf.voteCount > len(rf.peers)/2 {
                            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+1, rf.commitIndex)
        entriesToApply := append([]LogEntry{}, rf.log[rf.getRelativeLogIndex(rf.lastApplied+1):rf.getRelativeLogIndex(rf.commitIndex+1)]...)

        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+1, entriesToApply)
    }
}
func min(x, y int) int {
    if x < y {
        return x
    } else {
        return y
    }
}

//3B
func (rf *Raft) ReplaceLogWithSnapshot(appliedIndex int, kvSnapshot []byte) {
    rf.mu.Lock()
    defer rf.mu.Unlock()
    if appliedIndex <= rf.snapshottedIndex {
        return
    }
    // truncate log, keep snapshottedIndex as a guard at rf.log[0]
    // because it must be committed and applied
    rf.log = rf.log[rf.getRelativeLogIndex(appliedIndex):]
    rf.snapshottedIndex = appliedIndex
    rf.persister.SaveStateAndSnapshot(rf.encodeRaftState(), kvSnapshot)

    // update for other nodes
    for i := range rf.peers {
        if i == rf.me {
            continue
        }
        go rf.syncSnapshotWith(i)
    }
}

// invoke by Leader to sync snapshot with one follower
func (rf *Raft) syncSnapshotWith(server int) {
    rf.mu.Lock()
    if rf.state != Leader {
        rf.mu.Unlock()
        return
    }
    args := InstallSnapshotArgs{
        Term:              rf.currentTerm,
        LeaderId:          rf.me,
        LastIncludedIndex: rf.snapshottedIndex,
        LastIncludedTerm:  rf.log[0].Term,
        Data:              rf.persister.ReadSnapshot(),
    }
    DPrintf("%v sync snapshot with server %d for index %d, last snapshotted = %d",
        rf, server, args.LastIncludedIndex, rf.snapshottedIndex)
    rf.mu.Unlock()

    var reply InstallSnapshotReply

    if rf.sendInstallSnapshot(server, &args, &reply) {
        rf.mu.Lock()
        if reply.Term > rf.currentTerm {
            rf.currentTerm = reply.Term
            rf.switchStateTo(Follower)
            rf.persist()
        } else {
            if rf.matchIndex[server] < args.LastIncludedIndex {
                rf.matchIndex[server] = args.LastIncludedIndex
            }
            rf.nextIndex[server] = rf.matchIndex[server] + 1
        }
        rf.mu.Unlock()
    }
}

func (rf *Raft) getRelativeLogIndex(index int) int {
    // index of rf.log
    return index - rf.snapshottedIndex
}

func (rf *Raft) getAbsoluteLogIndex(index int) int {
    // index of log including snapshotted ones
    return index + rf.snapshottedIndex
}

type InstallSnapshotArgs struct {
    // do not need to implement "chunk"
    // remove "offset" and "done"
    Term              int    // 3B
    LeaderId          int    // 3B
    LastIncludedIndex int    // 3B
    LastIncludedTerm  int    // 3B
    Data              []byte // 3B
}

type InstallSnapshotReply struct {
    Term int // 3B
}

func (rf *Raft) InstallSnapshot(args *InstallSnapshotArgs, reply *InstallSnapshotReply) {
    rf.mu.Lock()
    defer rf.mu.Unlock()
    // we do not need to call rf.persist() in this function
    // because rf.persister.SaveStateAndSnapshot() is called

    reply.Term = rf.currentTerm
    if args.Term < rf.currentTerm || args.LastIncludedIndex < rf.snapshottedIndex {
        return
    }

    if args.Term > rf.currentTerm {
        rf.currentTerm = args.Term
        rf.switchStateTo(Follower)
        // do not return here.
    }

    // step 2, 3, 4 is skipped because we simplify the "offset"

    // 6. if existing log entry has same index and term with
    //    last log entry in snapshot, retain log entries following it
    lastIncludedRelativeIndex := rf.getRelativeLogIndex(args.LastIncludedIndex)
    if len(rf.log) > lastIncludedRelativeIndex &&
        rf.log[lastIncludedRelativeIndex].Term == args.LastIncludedTerm {
        rf.log = rf.log[lastIncludedRelativeIndex:]
    } else {
        // 7. discard entire log
        rf.log = []LogEntry{{Term: args.LastIncludedTerm, Command: nil}}
    }
    // 5. save snapshot file, discard any existing snapshot
    rf.snapshottedIndex = args.LastIncludedIndex
    // IMPORTANT: update commitIndex and lastApplied because after sync snapshot,
    //                         it has at least applied all logs before snapshottedIndex
    if rf.commitIndex < rf.snapshottedIndex {
        rf.commitIndex = rf.snapshottedIndex
    }
    if rf.lastApplied < rf.snapshottedIndex {
        rf.lastApplied = rf.snapshottedIndex
    }

    rf.persister.SaveStateAndSnapshot(rf.encodeRaftState(), args.Data)

    if rf.lastApplied > rf.snapshottedIndex {
        // snapshot is elder than kv's db
        // if we install snapshot on kvserver, linearizability will break
        return
    }

    installSnapshotCommand := ApplyMsg{
        CommandIndex: rf.snapshottedIndex,
        Command:      "InstallSnapshot",
        CommandValid: false,
        CommandData:  rf.persister.ReadSnapshot(),
    }
    go func(msg ApplyMsg) {
        rf.applyCh <- msg
    }(installSnapshotCommand)
}

func (rf *Raft) sendInstallSnapshot(server int, args *InstallSnapshotArgs, reply *InstallSnapshotReply) bool {
    ok := rf.peers[server].Call("Raft.InstallSnapshot", args, reply)
    return ok
}

server.go

package raftkv

import (
    "bytes"
    "labgob"
    "labrpc"
    "log"
    "raft"
    "sync"
    "time"
)

const Debug = 0

func DPrintf(format string, a ...interface{}) (n int, err error) {
    if Debug > 0 {
        log.Printf(format, a...)
    }
    return
}

type Op struct {
    // Your definitions here.
    // Field names must start with capital letters,
    // otherwise RPC will break.
    Key   string
    Value string
    Name  string

    ClientId  int64
    RequestId int
}

type KVServer struct {
    mu      sync.Mutex
    me      int
    rf      *raft.Raft
    applyCh chan raft.ApplyMsg

    maxraftstate int // snapshot if log grows this big

    // Your definitions here.
    db                   map[string]string         // 3A
    dispatcher           map[int]chan Notification // 3A
    lastAppliedRequestId map[int64]int             // 3A

    appliedRaftLogIndex int // 3B
}

// 3B
func (kv *KVServer) shouldTakeSnapshot() bool {
    if kv.maxraftstate == -1 {
        return false
    }

    if kv.rf.GetRaftStateSize() >= kv.maxraftstate {
        return true
    }
    return false
}

func (kv *KVServer) takeSnapshot() {
    w := new(bytes.Buffer)
    e := labgob.NewEncoder(w)
    kv.mu.Lock()
    e.Encode(kv.db)
    e.Encode(kv.lastAppliedRequestId)
    appliedRaftLogIndex := kv.appliedRaftLogIndex
    kv.mu.Unlock()

    kv.rf.ReplaceLogWithSnapshot(appliedRaftLogIndex, w.Bytes())
}

//3A
type Notification struct {
    ClientId  int64
    RequestId int
}

func (kv *KVServer) Get(args *GetArgs, reply *GetReply) {
    // Your code here.
    op := Op{
        Key:       args.Key,
        Name:      "Get",
        ClientId:  args.ClientId,
        RequestId: args.RequestId,
    }

    // wait for being applied
    // or leader changed (log is overrided, and never gets applied)
    reply.WrongLeader = kv.waitApplying(op, 500*time.Millisecond)

    if reply.WrongLeader == false {
        kv.mu.Lock()
        value, ok := kv.db[args.Key]
        kv.mu.Unlock()
        if ok {
            reply.Value = value
            return
        }
        // not found
        reply.Err = ErrNoKey
    }
}

func (kv *KVServer) PutAppend(args *PutAppendArgs, reply *PutAppendReply) {
    // Your code here.
    op := Op{
        Key:       args.Key,
        Value:     args.Value,
        Name:      args.Op,
        ClientId:  args.ClientId,
        RequestId: args.RequestId,
    }

    // wait for being applied
    // or leader changed (log is overrided, and never gets applied)
    reply.WrongLeader = kv.waitApplying(op, 500*time.Millisecond)
}

//
// the tester calls Kill() when a KVServer 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 (kv *KVServer) Kill() {
    kv.rf.Kill()
    // Your code here, if desired.
}

//
// servers[] contains the ports of the set of
// servers that will cooperate via Raft to
// form the fault-tolerant key/value service.
// me is the index of the current server in servers[].
// the k/v server should store snapshots through the underlying Raft
// implementation, which should call persister.SaveStateAndSnapshot() to
// atomically save the Raft state along with the snapshot.
// the k/v server should snapshot when Raft's saved state exceeds maxraftstate bytes,
// in order to allow Raft to garbage-collect its log. if maxraftstate is -1,
// you don't need to snapshot.
// StartKVServer() must return quickly, so it should start goroutines
// for any long-running work.
//
func StartKVServer(servers []*labrpc.ClientEnd, me int, persister *raft.Persister, maxraftstate int) *KVServer {
    // call labgob.Register on structures you want
    // Go's RPC library to marshall/unmarshall.
    labgob.Register(Op{})

    kv := new(KVServer)
    kv.me = me
    kv.maxraftstate = maxraftstate

    // You may need initialization code here.
    kv.db = make(map[string]string)
    kv.dispatcher = make(map[int]chan Notification)
    kv.lastAppliedRequestId = make(map[int64]int)

    kv.applyCh = make(chan raft.ApplyMsg)
    kv.rf = raft.Make(servers, me, persister, kv.applyCh)

    // 3B: recover from snapshot
    snapshot := persister.ReadSnapshot()
    kv.installSnapshot(snapshot)

    // You may need initialization code here.
    go func() {
        for msg := range kv.applyCh {
            if msg.CommandValid == false {
                //3B
                switch msg.Command.(string) {
                case "InstallSnapshot":
                    kv.installSnapshot(msg.CommandData)
                }
                continue
            }

            op := msg.Command.(Op)
            DPrintf("kvserver %d start applying command %s at index %d, request id %d, client id %d",
                kv.me, op.Name, msg.CommandIndex, op.RequestId, op.ClientId)
            kv.mu.Lock()
            if kv.isDuplicateRequest(op.ClientId, op.RequestId) {
                kv.mu.Unlock()
                continue
            }
            switch op.Name {
            case "Put":
                kv.db[op.Key] = op.Value
            case "Append":
                kv.db[op.Key] += op.Value
                // Get() does not need to modify db, skip
            }
            kv.lastAppliedRequestId[op.ClientId] = op.RequestId
            // 3B
            kv.appliedRaftLogIndex = msg.CommandIndex

            if ch, ok := kv.dispatcher[msg.CommandIndex]; ok {
                notify := Notification{
                    ClientId:  op.ClientId,
                    RequestId: op.RequestId,
                }
                ch <- notify
            }

            kv.mu.Unlock()
            DPrintf("kvserver %d applied command %s at index %d, request id %d, client id %d",
                kv.me, op.Name, msg.CommandIndex, op.RequestId, op.ClientId)
        }
    }()

    return kv
}

// should be called with lock
func (kv *KVServer) isDuplicateRequest(clientId int64, requestId int) bool {
    appliedRequestId, ok := kv.lastAppliedRequestId[clientId]
    if ok == false || requestId > appliedRequestId {
        return false
    }
    return true
}

func (kv *KVServer) waitApplying(op Op, timeout time.Duration) bool {
    // return common part of GetReply and PutAppendReply
    // i.e., WrongLeader
    index, _, isLeader := kv.rf.Start(op)
    if isLeader == false {
        return true
    }

    // 3B
    if kv.shouldTakeSnapshot() {
        kv.takeSnapshot()
    }

    var wrongLeader bool

    kv.mu.Lock()
    if _, ok := kv.dispatcher[index]; !ok {
        kv.dispatcher[index] = make(chan Notification, 1)
    }
    ch := kv.dispatcher[index]
    kv.mu.Unlock()
    select {
    case notify := <-ch:
        if notify.ClientId != op.ClientId || notify.RequestId != op.RequestId {
            // leader has changed
            wrongLeader = true
        } else {
            wrongLeader = false
        }

    case <-time.After(timeout):
        kv.mu.Lock()
        if kv.isDuplicateRequest(op.ClientId, op.RequestId) {
            wrongLeader = false
        } else {
            wrongLeader = true
        }
        kv.mu.Unlock()
    }
    DPrintf("kvserver %d got %s() RPC, insert op %+v at %d, reply WrongLeader = %v",
        kv.me, op.Name, op, index, wrongLeader)

    kv.mu.Lock()
    delete(kv.dispatcher, index)
    kv.mu.Unlock()
    return wrongLeader
}

// 3B
func (kv *KVServer) installSnapshot(snapshot []byte) {
    kv.mu.Lock()
    defer kv.mu.Unlock()
    if snapshot != nil {
        r := bytes.NewBuffer(snapshot)
        d := labgob.NewDecoder(r)
        if d.Decode(&kv.db) != nil ||
            d.Decode(&kv.lastAppliedRequestId) != nil {
            DPrintf("kvserver %d fails to recover from snapshot", kv.me)
        }
    }
}

go test  -run 3B

 go test -race -run 3B