golang[112]-raft理论与实践[3]-lab2a讲解

1、raft.go 的raft结构体 补充字段。 字段应该尽量与raft论文的Figure2接近。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
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[]
dead int32 // set by Kill()

// Your data here (2A, 2B, 2C).
// Look at the paper's Figure 2 for a description of what
// state a Raft server must maintain.
state int // follower, candidate or leader
resetTimer chan struct{} // for reset election timer
electionTimer *time.Timer // election timer
electionTimeout time.Duration // 400~800ms
heartbeatInterval time.Duration // 100ms

CurrentTerm int // Persisted before responding to RPCs
VotedFor int // Persisted before responding to RPCs
Logs []LogEntry // Persisted before responding to RPCs
commitCond *sync.Cond // for commitIndex update
//newEntryCond []*sync.Cond // for new log entry
commitIndex int // Volatile state on all servers
lastApplied int // Volatile state on all servers
nextIndex []int // Leader only, reinitialized after election
matchIndex []int // Leader only, reinitialized after election
applyCh chan ApplyMsg // outgoing channel to service
shutdownCh chan struct{} // shutdown channel, shut raft instance gracefully
}

获取当前raft节点的term与状态

1
2
3
4
5
6
7
8
9
10
11
12
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
}

2、填充RequestVoteArgs和RequestVoteReply结构。

1
2
3
4
5
6
7
8
9
10
11
12
13
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
LastLogTerm int // term of candidate's last log entry
}

type RequestVoteReply struct {
// Your data here (2A).
CurrentTerm int // currentTerm, for candidate to update itself
VoteGranted bool // true means candidate received vote
}

实现RPC方法RequestVote

1、获取当前节点的log个数,以及最后一个log的term 确定当前节点的term。
2、如果调用节点的term小于当前节点,返回当前term,并且不为其投票。
3、如果调用节点的term大于当前节点,修改当前节点的term,当前节点转为follower.
4、如果调用节点的term大于当前节点,或者等于当前节点term并且调用节点的log个数大于等于当前节点的log,则为调用节点投票。
5、投票后重置当前节点的选举超时时间。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
func (rf *Raft) RequestVote(args *RequestVoteArgs, reply *RequestVoteReply) {
// Your code here (2A, 2B).
select {
case <-rf.shutdownCh:
DPrintf("[%d-%s]: peer %d is shutting down, reject RV rpc request.\n", rf.me, rf, rf.me)
return
default:
}

rf.mu.Lock()
defer rf.mu.Unlock()

lastLogIdx, lastLogTerm := rf.lastLogIndexAndTerm()

DPrintf("[%d-%s]: rpc RV, from peer: %d, arg term: %d, my term: %d (last log idx: %d->%d, term: %d->%d)\n", rf.me, rf, args.CandidateID, args.Term, rf.CurrentTerm, args.LastLogIndex,
lastLogIdx, args.LastLogTerm, lastLogTerm)

if args.Term < rf.CurrentTerm {
reply.CurrentTerm = rf.CurrentTerm
reply.VoteGranted = false
} else {
if args.Term > rf.CurrentTerm {
// convert to follower
rf.CurrentTerm = args.Term
rf.state = Follower
rf.VotedFor = -1
}

// if is null (follower) or itself is a candidate (or stale leader) with same term
if rf.VotedFor == -1 { //|| (rf.VotedFor == rf.me && !sameTerm) { //|| rf.votedFor == args.CandidateID {
// check whether candidate's log is at-least-as update
if (args.LastLogTerm == lastLogTerm && args.LastLogIndex >= lastLogIdx) ||
args.LastLogTerm > lastLogTerm {

rf.resetTimer <- struct{}{}

rf.state = Follower
rf.VotedFor = args.CandidateID
reply.VoteGranted = true

DPrintf("[%d-%s]: peer %d vote to peer %d (last log idx: %d->%d, term: %d->%d)\n",
rf.me, rf, rf.me, args.CandidateID, args.LastLogIndex, lastLogIdx, args.LastLogTerm, lastLogTerm)
}
}
}
}

修改make

除了一些基本的初始化过程,新开了一个goroutine。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
func Make(peers []*labrpc.ClientEnd, me int,
persister *Persister, applyCh chan ApplyMsg) *Raft {
rf := &Raft{}
rf.peers = peers
rf.persister = persister
rf.me = me
rf.applyCh = applyCh
// Your initialization code here (2A, 2B, 2C).
rf.state = Follower
rf.VotedFor = -1
rf.Logs = make([]LogEntry, 1) // first index is 1
rf.Logs[0] = LogEntry{ // placeholder
Term: 0,
Command: nil,
}
rf.nextIndex = make([]int, len(peers))
rf.matchIndex = make([]int, len(peers))

rf.electionTimeout = time.Millisecond * time.Duration(400+rand.Intn(100)*4)
rf.electionTimer = time.NewTimer(rf.electionTimeout)
rf.resetTimer = make(chan struct{})
rf.shutdownCh = make(chan struct{}) // shutdown raft gracefully
rf.commitCond = sync.NewCond(&rf.mu) // commitCh, a distinct goroutine
rf.heartbeatInterval = time.Millisecond * 40 // small enough, not too small

// initialize from state persisted before a crash
rf.readPersist(persister.ReadRaftState())
go rf.electionDaemon() // kick off election
return rf
}

选举核心electionDaemon

除了shutdown,还有两个通道,一个是electionTimer,用于选举超时。
一个是resetTimer,用于重置选举超时。
注意time.reset是很难正确使用的。
一旦选举超时,调用go rf.canvassVotes()

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
// electionDaemon
func (rf *Raft) electionDaemon() {
for {
select {
case <-rf.shutdownCh:
DPrintf("[%d-%s]: peer %d is shutting down electionDaemon.\n", rf.me, rf, rf.me)
return
case <-rf.resetTimer:
if !rf.electionTimer.Stop() {
<-rf.electionTimer.C
}
rf.electionTimer.Reset(rf.electionTimeout)
case <-rf.electionTimer.C:
rf.mu.Lock()
DPrintf("[%d-%s]: peer %d election timeout, issue election @ term %d\n", rf.me, rf, rf.me, rf.CurrentTerm)
rf.mu.Unlock()
go rf.canvassVotes()
rf.electionTimer.Reset(rf.electionTimeout)
}
}
}

拉票

replyHandler是进行请求返回后的处理。
当前节点为了成为leader,会调用每一个节点的RequestVote方法。
如果返回过来的term大于当前term,那么当前节点变为follower,重置选举超时时间。

否则,如果收到了超过一半节点的投票,那么其变为了leader,并立即给其他节点发送心跳检测。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
// canvassVotes issues RequestVote RPC
func (rf *Raft) canvassVotes() {
var voteArgs RequestVoteArgs
rf.fillRequestVoteArgs(&voteArgs)
peers := len(rf.peers)

var votes = 1
replyHandler := func(reply *RequestVoteReply) {
rf.mu.Lock()
defer rf.mu.Unlock()
if rf.state == Candidate {
if reply.CurrentTerm > voteArgs.Term {
rf.CurrentTerm = reply.CurrentTerm
rf.turnToFollow()
//rf.persist()
rf.resetTimer <- struct{}{} // reset timer
return
}
if reply.VoteGranted {
if votes == peers/2 {
rf.state = Leader
rf.resetOnElection() // reset leader state
go rf.heartbeatDaemon() // new leader, start heartbeat daemon
DPrintf("[%d-%s]: peer %d become new leader.\n", rf.me, rf, rf.me)
return
}
votes++
}
}
}
for i := 0; i < peers; i++ {
if i != rf.me {
go func(n int) {
var reply RequestVoteReply
if rf.sendRequestVote(n, &voteArgs, &reply) {
replyHandler(&reply)
}
}(i)
}
}
}

心跳检测

1、leader调用每一个节点的AppendEntries方法。
2、如果当前节点大于调用节点,那么AppendEntries失败。否则,修改当前的term为最大。
3、如果当前节点是leader,始终将其变为follower(为了让leader稳定)
4、将当前节点投票给调用者(对于落后的节点)。
5、重置当前节点的超时时间。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
func (rf *Raft) heartbeatDaemon() {
for {
if _, isLeader := rf.GetState(); !isLeader {
return
}
// reset leader's election timer
rf.resetTimer <- struct{}{}

select {
case <-rf.shutdownCh:
return
default:
for i := 0; i < len(rf.peers); i++ {
if i != rf.me {
go rf.consistencyCheck(i) // routine heartbeat
}
}
}
time.Sleep(rf.heartbeatInterval)
}
}

func (rf *Raft) consistencyCheck(n int) {
rf.mu.Lock()
defer rf.mu.Unlock()
pre := rf.nextIndex[n] - 1
var args = AppendEntriesArgs{
Term: rf.CurrentTerm,
LeaderID: rf.me,
PrevLogIndex: pre,
PrevLogTerm: rf.Logs[pre].Term,
Entries: nil,
LeaderCommit: rf.commitIndex,
}
go func() {
DPrintf("[%d-%s]: consistency Check to peer %d.\n", rf.me, rf, n)
var reply AppendEntriesReply
if rf.sendAppendEntries(n, &args, &reply) {
rf.consistencyCheckReplyHandler(n, &reply)
}
}()
}
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
func (rf *Raft) AppendEntries(args *AppendEntriesArgs, reply *AppendEntriesReply) {
select {
case <-rf.shutdownCh:
DPrintf("[%d-%s]: peer %d is shutting down, reject AE rpc request.\n", rf.me, rf, rf.me)
return
default:
}

DPrintf("[%d-%s]: rpc AE, from peer: %d, term: %d\n", rf.me, rf, args.LeaderID, args.Term)
rf.mu.Lock()
defer rf.mu.Unlock()

if args.Term < rf.CurrentTerm {
//DPrintf("[%d-%s]: AE failed from leader %d. (heartbeat: leader's term < follower's term (%d < %d))\n",
// rf.me, rf, args.LeaderID, args.Term, rf.currentTerm)
reply.CurrentTerm = rf.CurrentTerm
reply.Success = false
return
}
if rf.CurrentTerm < args.Term {
rf.CurrentTerm = args.Term
}

// for stale leader
if rf.state == Leader {
rf.turnToFollow()
}
// for straggler (follower)
if rf.VotedFor != args.LeaderID {
rf.VotedFor = args.LeaderID
}

// valid AE, reset election timer
// if the node recieve heartbeat. then it will reset the election timeout
rf.resetTimer <- struct{}{}

reply.Success = true
reply.CurrentTerm = rf.CurrentTerm
return
}

处理心跳检测返回

如果心跳检测失败了,那么变为follower,重置选举超时。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
// n: which follower
func (rf *Raft) consistencyCheckReplyHandler(n int, reply *AppendEntriesReply) {
rf.mu.Lock()
defer rf.mu.Unlock()

if rf.state != Leader {
return
}
if reply.Success {

} else {
// found a new leader? turn to follower
if rf.state == Leader && reply.CurrentTerm > rf.CurrentTerm {
rf.turnToFollow()
rf.resetTimer <- struct{}{}
DPrintf("[%d-%s]: leader %d found new term (heartbeat resp from peer %d), turn to follower.",
rf.me, rf, rf.me, n)
return
}
}
}