选举

etcd的选举则需要在我们熟悉它的一系列基本概念后，调动我们充分的想象力：

1、MVCC，key存在版本属性，没被创建时版本号为0；

2、CAS操作，结合MVCC，可以实现竞选逻辑，if(version == 0) set(key,value),通过原子操作，确保只有一台机器能set成功；

3、Lease租约，可以对key绑定一个租约，租约到期时没预约，这个key就会被回收；

4、Watch监听，监听key的变化事件，如果key被删除，则重新发起竞选。

至此，etcd选举的逻辑大体清晰了，etcdctl ctlv3里已经有master选举的实现了，下面针对这部分代码进行简单注释，在最后参考这部分代码实现自己的选举逻辑。

  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
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254


// Copyright 2016 The etcd Authors
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

package concurrency

import (
	"context"
	"errors"
	"fmt"

	v3 "github.com/coreos/etcd/clientv3"
	pb "github.com/coreos/etcd/etcdserver/etcdserverpb"
	"github.com/coreos/etcd/mvcc/mvccpb"
)

var (
	ErrElectionNotLeader = errors.New("election: not leader")
	ErrElectionNoLeader  = errors.New("election: no leader")
)

type Election struct {
	session *Session

	keyPrefix string

	leaderKey     string
	leaderRev     int64
	leaderSession *Session
	hdr           *pb.ResponseHeader
}

// NewElection returns a new election on a given key prefix.
func NewElection(s *Session, pfx string) *Election {
	return &Election{session: s, keyPrefix: pfx + "/"}
}

// ResumeElection initializes an election with a known leader.
func ResumeElection(s *Session, pfx string, leaderKey string, leaderRev int64) *Election {
	return &Election{
		session:       s,
		leaderKey:     leaderKey,
		leaderRev:     leaderRev,
		leaderSession: s,
	}
}
/*
 * etcd的选举是在相应的prefix path下面创建key，该key绑定了lease并根据lease id进行命名，
 * key创建后就有revision号，这样使得在prefix path下的key也都是按revision有序
 */
// Campaign puts a value as eligible for the election. It blocks until
// it is elected, an error occurs, or the context is cancelled.
func (e *Election) Campaign(ctx context.Context, val string) error {
	s := e.session
	client := e.session.Client()
	//真正创建的key名为：prefix + lease id
	k := fmt.Sprintf("%s%x", e.keyPrefix, s.Lease())
	//Txn：transaction，依靠Txn进行创建key的CAS操作，当key不存在时才会成功创建
	txn := client.Txn(ctx).If(v3.Compare(v3.CreateRevision(k), "=", 0))
	txn = txn.Then(v3.OpPut(k, val, v3.WithLease(s.Lease())))
	txn = txn.Else(v3.OpGet(k))
	resp, err := txn.Commit()
	if err != nil {
		return err
	}
	e.leaderKey, e.leaderRev, e.leaderSession = k, resp.Header.Revision, s
	//如果key已存在
	if !resp.Succeeded {
		//获取当前的kv
		kv := resp.Responses[0].GetResponseRange().Kvs[0]
		e.leaderRev = kv.CreateRevision
		//当key的value与当前value不等时，如果自己为leader，则不用重新执行选举直接设置value；
		//否则报错。
		if string(kv.Value) != val {
			//用新值更新领导者的旧值
			if err = e.Proclaim(ctx, val); err != nil {
				e.Resign(ctx)
				return err
			}
		}
	}
	//一直阻塞，直到确认自己的create revision为当前path中最小，从而确认自己当选为leader
	_, err = waitDeletes(ctx, client, e.keyPrefix, e.leaderRev-1)
	if err != nil {
		// clean up in case of context cancel
		select {
		case <-ctx.Done():
			e.Resign(client.Ctx())
		default:
			e.leaderSession = nil
		}
		return err
	}
	e.hdr = resp.Header

	return nil
}

// Proclaim lets the leader announce a new value without another election.
func (e *Election) Proclaim(ctx context.Context, val string) error {
	if e.leaderSession == nil {
		return ErrElectionNotLeader
	}
	client := e.session.Client()
	cmp := v3.Compare(v3.CreateRevision(e.leaderKey), "=", e.leaderRev)
	txn := client.Txn(ctx).If(cmp)
	txn = txn.Then(v3.OpPut(e.leaderKey, val, v3.WithLease(e.leaderSession.Lease())))
	tresp, terr := txn.Commit()
	if terr != nil {
		return terr
	}
	if !tresp.Succeeded {
		e.leaderKey = ""
		return ErrElectionNotLeader
	}

	e.hdr = tresp.Header
	return nil
}

// Resign lets a leader start a new election.
func (e *Election) Resign(ctx context.Context) (err error) {
	if e.leaderSession == nil {
		return nil
	}
	client := e.session.Client()
	cmp := v3.Compare(v3.CreateRevision(e.leaderKey), "=", e.leaderRev)
	resp, err := client.Txn(ctx).If(cmp).Then(v3.OpDelete(e.leaderKey)).Commit()
	if err == nil {
		e.hdr = resp.Header
	}
	e.leaderKey = ""
	e.leaderSession = nil
	return err
}

// Leader returns the leader value for the current election.
func (e *Election) Leader(ctx context.Context) (*v3.GetResponse, error) {
	client := e.session.Client()
	resp, err := client.Get(ctx, e.keyPrefix, v3.WithFirstCreate()...)
	if err != nil {
		return nil, err
	} else if len(resp.Kvs) == 0 {
		// no leader currently elected
		return nil, ErrElectionNoLeader
	}
	return resp, nil
}

// Observe returns a channel that reliably observes ordered leader proposals
// as GetResponse values on every current elected leader key. It will not
// necessarily fetch all historical leader updates, but will always post the
// most recent leader value.
//
// The channel closes when the context is canceled or the underlying watcher
// is otherwise disrupted.
func (e *Election) Observe(ctx context.Context) <-chan v3.GetResponse {
	retc := make(chan v3.GetResponse)
	go e.observe(ctx, retc)
	return retc
}

func (e *Election) observe(ctx context.Context, ch chan<- v3.GetResponse) {
	client := e.session.Client()

	defer close(ch)
	for {
		resp, err := client.Get(ctx, e.keyPrefix, v3.WithFirstCreate()...)
		if err != nil {
			return
		}

		var kv *mvccpb.KeyValue
		var hdr *pb.ResponseHeader

		if len(resp.Kvs) == 0 {
			cctx, cancel := context.WithCancel(ctx)
			// wait for first key put on prefix
			opts := []v3.OpOption{v3.WithRev(resp.Header.Revision), v3.WithPrefix()}
			wch := client.Watch(cctx, e.keyPrefix, opts...)
			for kv == nil {
				wr, ok := <-wch
				if !ok || wr.Err() != nil {
					cancel()
					return
				}
				// only accept puts; a delete will make observe() spin
				for _, ev := range wr.Events {
					if ev.Type == mvccpb.PUT {
						hdr, kv = &wr.Header, ev.Kv
						// may have multiple revs; hdr.rev = the last rev
						// set to kv's rev in case batch has multiple Puts
						hdr.Revision = kv.ModRevision
						break
					}
				}
			}
			cancel()
		} else {
			hdr, kv = resp.Header, resp.Kvs[0]
		}

		select {
		case ch <- v3.GetResponse{Header: hdr, Kvs: []*mvccpb.KeyValue{kv}}:
		case <-ctx.Done():
			return
		}

		cctx, cancel := context.WithCancel(ctx)
		wch := client.Watch(cctx, string(kv.Key), v3.WithRev(hdr.Revision+1))
		keyDeleted := false
		for !keyDeleted {
			wr, ok := <-wch
			if !ok {
				cancel()
				return
			}
			for _, ev := range wr.Events {
				if ev.Type == mvccpb.DELETE {
					keyDeleted = true
					break
				}
				resp.Header = &wr.Header
				resp.Kvs = []*mvccpb.KeyValue{ev.Kv}
				select {
				case ch <- *resp:
				case <-cctx.Done():
					cancel()
					return
				}
			}
		}
		cancel()
	}
}

// Key returns the leader key if elected, empty string otherwise.
func (e *Election) Key() string { return e.leaderKey }

// Rev returns the leader key's creation revision, if elected.
func (e *Election) Rev() int64 { return e.leaderRev }

// Header is the response header from the last successful election proposal.
func (e *Election) Header() *pb.ResponseHeader { return e.hdr }

最后给出一个Master选举的demo，模拟的场景：每5s执行一次crontab，且只有leader节点能执行crontab；lease租约有效期设为15s，leader节点异常，最长等待15s，集群会产生新的leader执行crontab。

 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
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72


package main

import (
	"context"
	"fmt"
	"go.etcd.io/etcd/v3/clientv3"
	"go.etcd.io/etcd/v3/clientv3/concurrency"
	"log"
	"time"
)

const prefix = "/election-demo"
const prop = "local"

var leaderFlag bool

func main() {
	endpoints := []string{"127.0.0.1:2379"}
	donec := make(chan struct{})

	cli, err := clientv3.New(clientv3.Config{Endpoints: endpoints})
	if err != nil {
		log.Fatal(err)
	}
	defer cli.Close()

	go campaign(cli, prefix, prop)

	go func() {
		ticker := time.NewTicker(time.Duration(5) * time.Second)
		for {
			select {
			case <-ticker.C:
				doCrontab()
			}
		}
	}()

	<-donec
}

func campaign(c *clientv3.Client, election string, prop string) {
	for {
		s, err := concurrency.NewSession(c, concurrency.WithTTL(15))
		if err != nil {
			fmt.Println(err)
			continue
		}
		e := concurrency.NewElection(s, election)
		ctx := context.TODO()

		if err = e.Campaign(ctx, prop); err != nil {
			fmt.Println(err)
			continue
		}

		fmt.Println("elect: success")
		leaderFlag = true

		select {
		case <-s.Done():
			leaderFlag = false
			fmt.Println("elect: expired")
		}
	}
}

func doCrontab() {
	if leaderFlag == true {
		fmt.Println("doCrontab")
	}
}

参考:https://cloud.tencent.com/developer/article/1458456

用eTcd实现选举

文章目录

选举