raft理論與實踐[3]-lab2a講解
lab2a實驗講解
一、閱讀raft論文
二、閱讀raft理論與實踐[1]-理論篇
三、閱讀raft理論與實踐[2]-lab2a
四、查看我寫的這篇文章： 模擬RPC遠程過程調用node
一、raft.go 的raft結構體 補充字段。 字段應該儘可能與raft論文的Figure2接近。golang
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
}
獲取當前raft節點的term與狀態
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
}
二、填充RequestVoteArgs和RequestVoteReply結構。
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
}
實現RPC方法RequestVote
一、獲取當前節點的log個數，以及最後一個log的term 肯定當前節點的term。canvas
二、若是調用節點的term小於當前節點，返回當前term，而且不爲其投票。app
三、若是調用節點的term大於當前節點，修改當前節點的term，當前節點轉爲follower.ide
四、若是調用節點的term大於當前節點，或者等於當前節點term而且調用節點的log個數大於等於當前節點的log，則爲調用節點投票。this
五、投票後重置當前節點的選舉超時時間。code
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過程,新開了一個goroutine。server
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
}
選舉核心electionDaemon
除了shutdown，還有兩個通道，一個是electionTimer，用於選舉超時。ip
一個是resetTimer,用於重置選舉超時。ci
注意time.reset是很難正確使用的。
一旦選舉超時，調用go rf.canvassVotes()
// 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是進行請求返回後的處理。
當前節點爲了成爲leader，會調用每個節點的RequestVote方法。
若是返回過來的term大於當前term，那麼當前節點變爲follower,重置選舉超時時間。
不然，若是收到了超過一半節點的投票,那麼其變爲了leader，並當即給其餘節點發送心跳檢測。
// 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)
}
}
}
心跳檢測
一、leader調用每個節點的AppendEntries方法。
二、若是當前節點大於調用節點，那麼AppendEntries失敗。不然，修改當前的term爲最大。
三、若是當前節點是leader，始終將其變爲follower（爲了讓leader穩定）
四、將當前節點投票給調用者（對於落後的節點）。
五、重置當前節點的超時時間。
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)
}
}()
}
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
}
處理心跳檢測返回
若是心跳檢測失敗了，那麼變爲follower，重置選舉超時。
// 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
}
}
}
相關文章
- 1. raft理論與實踐[4]-lab2b
- 2. raft理論與實踐[2]-lab2a實驗說明
- 3. raft理論與實踐[5]-lab2c-持久化
- 4. 理解RESTful：理論與最佳實踐
- 5. SSH原理講解與實踐
- 6. HTTPS 理論詳解與實踐
- 7. word2vec理論與實踐
- 8. 冪等理論與實踐
- 9. C++理論與實踐
- 10. FM理論與實踐
- 更多相關文章...
- • Thymeleaf項目實踐 - Thymeleaf 教程
- • CAP理論是什麼？ - NoSQL教程
- • Java Agent入門實戰（三）-JVM Attach原理與使用
- • TiDB 在摩拜單車在線數據業務的應用和實踐
相關標籤/搜索
每日一句
-
每一个你不满意的现在，都有一个你没有努力的曾经。
歡迎關注本站公眾號,獲取更多信息
相關文章
- 1. raft理論與實踐[4]-lab2b
- 2. raft理論與實踐[2]-lab2a實驗說明
- 3. raft理論與實踐[5]-lab2c-持久化
- 4. 理解RESTful：理論與最佳實踐
- 5. SSH原理講解與實踐
- 6. HTTPS 理論詳解與實踐
- 7. word2vec理論與實踐
- 8. 冪等理論與實踐
- 9. C++理論與實踐
- 10. FM理論與實踐