election.go 6.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253
  1. // Copyright 2016 The etcd Authors
  2. //
  3. // Licensed under the Apache License, Version 2.0 (the "License");
  4. // you may not use this file except in compliance with the License.
  5. // You may obtain a copy of the License at
  6. //
  7. // http://www.apache.org/licenses/LICENSE-2.0
  8. //
  9. // Unless required by applicable law or agreed to in writing, software
  10. // distributed under the License is distributed on an "AS IS" BASIS,
  11. // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  12. // See the License for the specific language governing permissions and
  13. // limitations under the License.
  14. package concurrency
  15. import (
  16. "context"
  17. "errors"
  18. "fmt"
  19. v3 "go.etcd.io/etcd/clientv3"
  20. pb "go.etcd.io/etcd/etcdserver/etcdserverpb"
  21. "go.etcd.io/etcd/mvcc/mvccpb"
  22. )
  23. var (
  24. ErrElectionNotLeader = errors.New("election: not leader")
  25. ErrElectionNoLeader = errors.New("election: no leader")
  26. )
  27. type Election struct {
  28. session *Session
  29. keyPrefix string
  30. leaderKey string
  31. leaderRev int64
  32. leaderSession *Session
  33. hdr *pb.ResponseHeader
  34. }
  35. // NewElection returns a new election on a given key prefix.
  36. func NewElection(s *Session, pfx string) *Election {
  37. return &Election{session: s, keyPrefix: pfx + "/"}
  38. }
  39. // ResumeElection initializes an election with a known leader.
  40. func ResumeElection(s *Session, pfx string, leaderKey string, leaderRev int64) *Election {
  41. return &Election{
  42. session: s,
  43. leaderKey: leaderKey,
  44. leaderRev: leaderRev,
  45. leaderSession: s,
  46. }
  47. }
  48. // Campaign puts a value as eligible for the election on the prefix
  49. // key.
  50. // Multiple sessions can participate in the election for the
  51. // same prefix, but only one can be the leader at a time.
  52. //
  53. // If the context is 'context.TODO()/context.Background()', the Campaign
  54. // will continue to be blocked for other keys to be deleted, unless server
  55. // returns a non-recoverable error (e.g. ErrCompacted).
  56. // Otherwise, until the context is not cancelled or timed-out, Campaign will
  57. // continue to be blocked until it becomes the leader.
  58. func (e *Election) Campaign(ctx context.Context, val string) error {
  59. s := e.session
  60. client := e.session.Client()
  61. k := fmt.Sprintf("%s%x", e.keyPrefix, s.Lease())
  62. txn := client.Txn(ctx).If(v3.Compare(v3.CreateRevision(k), "=", 0))
  63. txn = txn.Then(v3.OpPut(k, val, v3.WithLease(s.Lease())))
  64. txn = txn.Else(v3.OpGet(k))
  65. resp, err := txn.Commit()
  66. if err != nil {
  67. return err
  68. }
  69. e.leaderKey, e.leaderRev, e.leaderSession = k, resp.Header.Revision, s
  70. if !resp.Succeeded {
  71. kv := resp.Responses[0].GetResponseRange().Kvs[0]
  72. e.leaderRev = kv.CreateRevision
  73. if string(kv.Value) != val {
  74. if err = e.Proclaim(ctx, val); err != nil {
  75. e.Resign(ctx)
  76. return err
  77. }
  78. }
  79. }
  80. _, err = waitDeletes(ctx, client, e.keyPrefix, e.leaderRev-1)
  81. if err != nil {
  82. // clean up in case of context cancel
  83. select {
  84. case <-ctx.Done():
  85. e.Resign(client.Ctx())
  86. default:
  87. e.leaderSession = nil
  88. }
  89. return err
  90. }
  91. e.hdr = resp.Header
  92. return nil
  93. }
  94. // Proclaim lets the leader announce a new value without another election.
  95. func (e *Election) Proclaim(ctx context.Context, val string) error {
  96. if e.leaderSession == nil {
  97. return ErrElectionNotLeader
  98. }
  99. client := e.session.Client()
  100. cmp := v3.Compare(v3.CreateRevision(e.leaderKey), "=", e.leaderRev)
  101. txn := client.Txn(ctx).If(cmp)
  102. txn = txn.Then(v3.OpPut(e.leaderKey, val, v3.WithLease(e.leaderSession.Lease())))
  103. tresp, terr := txn.Commit()
  104. if terr != nil {
  105. return terr
  106. }
  107. if !tresp.Succeeded {
  108. e.leaderKey = ""
  109. return ErrElectionNotLeader
  110. }
  111. e.hdr = tresp.Header
  112. return nil
  113. }
  114. // Resign lets a leader start a new election.
  115. func (e *Election) Resign(ctx context.Context) (err error) {
  116. if e.leaderSession == nil {
  117. return nil
  118. }
  119. client := e.session.Client()
  120. cmp := v3.Compare(v3.CreateRevision(e.leaderKey), "=", e.leaderRev)
  121. resp, err := client.Txn(ctx).If(cmp).Then(v3.OpDelete(e.leaderKey)).Commit()
  122. if err == nil {
  123. e.hdr = resp.Header
  124. }
  125. e.leaderKey = ""
  126. e.leaderSession = nil
  127. return err
  128. }
  129. // Leader returns the leader value for the current election.
  130. func (e *Election) Leader(ctx context.Context) (*v3.GetResponse, error) {
  131. client := e.session.Client()
  132. resp, err := client.Get(ctx, e.keyPrefix, v3.WithFirstCreate()...)
  133. if err != nil {
  134. return nil, err
  135. } else if len(resp.Kvs) == 0 {
  136. // no leader currently elected
  137. return nil, ErrElectionNoLeader
  138. }
  139. return resp, nil
  140. }
  141. // Observe returns a channel that reliably observes ordered leader proposals
  142. // as GetResponse values on every current elected leader key. It will not
  143. // necessarily fetch all historical leader updates, but will always post the
  144. // most recent leader value.
  145. //
  146. // The channel closes when the context is canceled or the underlying watcher
  147. // is otherwise disrupted.
  148. func (e *Election) Observe(ctx context.Context) <-chan v3.GetResponse {
  149. retc := make(chan v3.GetResponse)
  150. go e.observe(ctx, retc)
  151. return retc
  152. }
  153. func (e *Election) observe(ctx context.Context, ch chan<- v3.GetResponse) {
  154. client := e.session.Client()
  155. defer close(ch)
  156. for {
  157. resp, err := client.Get(ctx, e.keyPrefix, v3.WithFirstCreate()...)
  158. if err != nil {
  159. return
  160. }
  161. var kv *mvccpb.KeyValue
  162. var hdr *pb.ResponseHeader
  163. if len(resp.Kvs) == 0 {
  164. cctx, cancel := context.WithCancel(ctx)
  165. // wait for first key put on prefix
  166. opts := []v3.OpOption{v3.WithRev(resp.Header.Revision), v3.WithPrefix()}
  167. wch := client.Watch(cctx, e.keyPrefix, opts...)
  168. for kv == nil {
  169. wr, ok := <-wch
  170. if !ok || wr.Err() != nil {
  171. cancel()
  172. return
  173. }
  174. // only accept puts; a delete will make observe() spin
  175. for _, ev := range wr.Events {
  176. if ev.Type == mvccpb.PUT {
  177. hdr, kv = &wr.Header, ev.Kv
  178. // may have multiple revs; hdr.rev = the last rev
  179. // set to kv's rev in case batch has multiple Puts
  180. hdr.Revision = kv.ModRevision
  181. break
  182. }
  183. }
  184. }
  185. cancel()
  186. } else {
  187. hdr, kv = resp.Header, resp.Kvs[0]
  188. }
  189. select {
  190. case ch <- v3.GetResponse{Header: hdr, Kvs: []*mvccpb.KeyValue{kv}}:
  191. case <-ctx.Done():
  192. return
  193. }
  194. cctx, cancel := context.WithCancel(ctx)
  195. wch := client.Watch(cctx, string(kv.Key), v3.WithRev(hdr.Revision+1))
  196. keyDeleted := false
  197. for !keyDeleted {
  198. wr, ok := <-wch
  199. if !ok {
  200. cancel()
  201. return
  202. }
  203. for _, ev := range wr.Events {
  204. if ev.Type == mvccpb.DELETE {
  205. keyDeleted = true
  206. break
  207. }
  208. resp.Header = &wr.Header
  209. resp.Kvs = []*mvccpb.KeyValue{ev.Kv}
  210. select {
  211. case ch <- *resp:
  212. case <-cctx.Done():
  213. cancel()
  214. return
  215. }
  216. }
  217. }
  218. cancel()
  219. }
  220. }
  221. // Key returns the leader key if elected, empty string otherwise.
  222. func (e *Election) Key() string { return e.leaderKey }
  223. // Rev returns the leader key's creation revision, if elected.
  224. func (e *Election) Rev() int64 { return e.leaderRev }
  225. // Header is the response header from the last successful election proposal.
  226. func (e *Election) Header() *pb.ResponseHeader { return e.hdr }