khenaidoo | ab1f7bd | 2019-11-14 14:00:27 -0500 | [diff] [blame] | 1 | // Copyright 2015 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 | |
| 15 | package rafthttp |
| 16 | |
| 17 | import ( |
| 18 | "context" |
| 19 | "net/http" |
| 20 | "sync" |
| 21 | "time" |
| 22 | |
| 23 | "go.etcd.io/etcd/etcdserver/api/snap" |
| 24 | stats "go.etcd.io/etcd/etcdserver/api/v2stats" |
| 25 | "go.etcd.io/etcd/pkg/logutil" |
| 26 | "go.etcd.io/etcd/pkg/transport" |
| 27 | "go.etcd.io/etcd/pkg/types" |
| 28 | "go.etcd.io/etcd/raft" |
| 29 | "go.etcd.io/etcd/raft/raftpb" |
| 30 | |
| 31 | "github.com/coreos/pkg/capnslog" |
| 32 | "github.com/xiang90/probing" |
| 33 | "go.uber.org/zap" |
| 34 | "golang.org/x/time/rate" |
| 35 | ) |
| 36 | |
| 37 | var plog = logutil.NewMergeLogger(capnslog.NewPackageLogger("go.etcd.io/etcd", "rafthttp")) |
| 38 | |
| 39 | type Raft interface { |
| 40 | Process(ctx context.Context, m raftpb.Message) error |
| 41 | IsIDRemoved(id uint64) bool |
| 42 | ReportUnreachable(id uint64) |
| 43 | ReportSnapshot(id uint64, status raft.SnapshotStatus) |
| 44 | } |
| 45 | |
| 46 | type Transporter interface { |
| 47 | // Start starts the given Transporter. |
| 48 | // Start MUST be called before calling other functions in the interface. |
| 49 | Start() error |
| 50 | // Handler returns the HTTP handler of the transporter. |
| 51 | // A transporter HTTP handler handles the HTTP requests |
| 52 | // from remote peers. |
| 53 | // The handler MUST be used to handle RaftPrefix(/raft) |
| 54 | // endpoint. |
| 55 | Handler() http.Handler |
| 56 | // Send sends out the given messages to the remote peers. |
| 57 | // Each message has a To field, which is an id that maps |
| 58 | // to an existing peer in the transport. |
| 59 | // If the id cannot be found in the transport, the message |
| 60 | // will be ignored. |
| 61 | Send(m []raftpb.Message) |
| 62 | // SendSnapshot sends out the given snapshot message to a remote peer. |
| 63 | // The behavior of SendSnapshot is similar to Send. |
| 64 | SendSnapshot(m snap.Message) |
| 65 | // AddRemote adds a remote with given peer urls into the transport. |
| 66 | // A remote helps newly joined member to catch up the progress of cluster, |
| 67 | // and will not be used after that. |
| 68 | // It is the caller's responsibility to ensure the urls are all valid, |
| 69 | // or it panics. |
| 70 | AddRemote(id types.ID, urls []string) |
| 71 | // AddPeer adds a peer with given peer urls into the transport. |
| 72 | // It is the caller's responsibility to ensure the urls are all valid, |
| 73 | // or it panics. |
| 74 | // Peer urls are used to connect to the remote peer. |
| 75 | AddPeer(id types.ID, urls []string) |
| 76 | // RemovePeer removes the peer with given id. |
| 77 | RemovePeer(id types.ID) |
| 78 | // RemoveAllPeers removes all the existing peers in the transport. |
| 79 | RemoveAllPeers() |
| 80 | // UpdatePeer updates the peer urls of the peer with the given id. |
| 81 | // It is the caller's responsibility to ensure the urls are all valid, |
| 82 | // or it panics. |
| 83 | UpdatePeer(id types.ID, urls []string) |
| 84 | // ActiveSince returns the time that the connection with the peer |
| 85 | // of the given id becomes active. |
| 86 | // If the connection is active since peer was added, it returns the adding time. |
| 87 | // If the connection is currently inactive, it returns zero time. |
| 88 | ActiveSince(id types.ID) time.Time |
| 89 | // ActivePeers returns the number of active peers. |
| 90 | ActivePeers() int |
| 91 | // Stop closes the connections and stops the transporter. |
| 92 | Stop() |
| 93 | } |
| 94 | |
| 95 | // Transport implements Transporter interface. It provides the functionality |
| 96 | // to send raft messages to peers, and receive raft messages from peers. |
| 97 | // User should call Handler method to get a handler to serve requests |
| 98 | // received from peerURLs. |
| 99 | // User needs to call Start before calling other functions, and call |
| 100 | // Stop when the Transport is no longer used. |
| 101 | type Transport struct { |
| 102 | Logger *zap.Logger |
| 103 | |
| 104 | DialTimeout time.Duration // maximum duration before timing out dial of the request |
| 105 | // DialRetryFrequency defines the frequency of streamReader dial retrial attempts; |
| 106 | // a distinct rate limiter is created per every peer (default value: 10 events/sec) |
| 107 | DialRetryFrequency rate.Limit |
| 108 | |
| 109 | TLSInfo transport.TLSInfo // TLS information used when creating connection |
| 110 | |
| 111 | ID types.ID // local member ID |
| 112 | URLs types.URLs // local peer URLs |
| 113 | ClusterID types.ID // raft cluster ID for request validation |
| 114 | Raft Raft // raft state machine, to which the Transport forwards received messages and reports status |
| 115 | Snapshotter *snap.Snapshotter |
| 116 | ServerStats *stats.ServerStats // used to record general transportation statistics |
| 117 | // used to record transportation statistics with followers when |
| 118 | // performing as leader in raft protocol |
| 119 | LeaderStats *stats.LeaderStats |
| 120 | // ErrorC is used to report detected critical errors, e.g., |
| 121 | // the member has been permanently removed from the cluster |
| 122 | // When an error is received from ErrorC, user should stop raft state |
| 123 | // machine and thus stop the Transport. |
| 124 | ErrorC chan error |
| 125 | |
| 126 | streamRt http.RoundTripper // roundTripper used by streams |
| 127 | pipelineRt http.RoundTripper // roundTripper used by pipelines |
| 128 | |
| 129 | mu sync.RWMutex // protect the remote and peer map |
| 130 | remotes map[types.ID]*remote // remotes map that helps newly joined member to catch up |
| 131 | peers map[types.ID]Peer // peers map |
| 132 | |
| 133 | pipelineProber probing.Prober |
| 134 | streamProber probing.Prober |
| 135 | } |
| 136 | |
| 137 | func (t *Transport) Start() error { |
| 138 | var err error |
| 139 | t.streamRt, err = newStreamRoundTripper(t.TLSInfo, t.DialTimeout) |
| 140 | if err != nil { |
| 141 | return err |
| 142 | } |
| 143 | t.pipelineRt, err = NewRoundTripper(t.TLSInfo, t.DialTimeout) |
| 144 | if err != nil { |
| 145 | return err |
| 146 | } |
| 147 | t.remotes = make(map[types.ID]*remote) |
| 148 | t.peers = make(map[types.ID]Peer) |
| 149 | t.pipelineProber = probing.NewProber(t.pipelineRt) |
| 150 | t.streamProber = probing.NewProber(t.streamRt) |
| 151 | |
| 152 | // If client didn't provide dial retry frequency, use the default |
| 153 | // (100ms backoff between attempts to create a new stream), |
| 154 | // so it doesn't bring too much overhead when retry. |
| 155 | if t.DialRetryFrequency == 0 { |
| 156 | t.DialRetryFrequency = rate.Every(100 * time.Millisecond) |
| 157 | } |
| 158 | return nil |
| 159 | } |
| 160 | |
| 161 | func (t *Transport) Handler() http.Handler { |
| 162 | pipelineHandler := newPipelineHandler(t, t.Raft, t.ClusterID) |
| 163 | streamHandler := newStreamHandler(t, t, t.Raft, t.ID, t.ClusterID) |
| 164 | snapHandler := newSnapshotHandler(t, t.Raft, t.Snapshotter, t.ClusterID) |
| 165 | mux := http.NewServeMux() |
| 166 | mux.Handle(RaftPrefix, pipelineHandler) |
| 167 | mux.Handle(RaftStreamPrefix+"/", streamHandler) |
| 168 | mux.Handle(RaftSnapshotPrefix, snapHandler) |
| 169 | mux.Handle(ProbingPrefix, probing.NewHandler()) |
| 170 | return mux |
| 171 | } |
| 172 | |
| 173 | func (t *Transport) Get(id types.ID) Peer { |
| 174 | t.mu.RLock() |
| 175 | defer t.mu.RUnlock() |
| 176 | return t.peers[id] |
| 177 | } |
| 178 | |
| 179 | func (t *Transport) Send(msgs []raftpb.Message) { |
| 180 | for _, m := range msgs { |
| 181 | if m.To == 0 { |
| 182 | // ignore intentionally dropped message |
| 183 | continue |
| 184 | } |
| 185 | to := types.ID(m.To) |
| 186 | |
| 187 | t.mu.RLock() |
| 188 | p, pok := t.peers[to] |
| 189 | g, rok := t.remotes[to] |
| 190 | t.mu.RUnlock() |
| 191 | |
| 192 | if pok { |
| 193 | if m.Type == raftpb.MsgApp { |
| 194 | t.ServerStats.SendAppendReq(m.Size()) |
| 195 | } |
| 196 | p.send(m) |
| 197 | continue |
| 198 | } |
| 199 | |
| 200 | if rok { |
| 201 | g.send(m) |
| 202 | continue |
| 203 | } |
| 204 | |
| 205 | if t.Logger != nil { |
| 206 | t.Logger.Debug( |
| 207 | "ignored message send request; unknown remote peer target", |
| 208 | zap.String("type", m.Type.String()), |
| 209 | zap.String("unknown-target-peer-id", to.String()), |
| 210 | ) |
| 211 | } else { |
| 212 | plog.Debugf("ignored message %s (sent to unknown peer %s)", m.Type, to) |
| 213 | } |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | func (t *Transport) Stop() { |
| 218 | t.mu.Lock() |
| 219 | defer t.mu.Unlock() |
| 220 | for _, r := range t.remotes { |
| 221 | r.stop() |
| 222 | } |
| 223 | for _, p := range t.peers { |
| 224 | p.stop() |
| 225 | } |
| 226 | t.pipelineProber.RemoveAll() |
| 227 | t.streamProber.RemoveAll() |
| 228 | if tr, ok := t.streamRt.(*http.Transport); ok { |
| 229 | tr.CloseIdleConnections() |
| 230 | } |
| 231 | if tr, ok := t.pipelineRt.(*http.Transport); ok { |
| 232 | tr.CloseIdleConnections() |
| 233 | } |
| 234 | t.peers = nil |
| 235 | t.remotes = nil |
| 236 | } |
| 237 | |
| 238 | // CutPeer drops messages to the specified peer. |
| 239 | func (t *Transport) CutPeer(id types.ID) { |
| 240 | t.mu.RLock() |
| 241 | p, pok := t.peers[id] |
| 242 | g, gok := t.remotes[id] |
| 243 | t.mu.RUnlock() |
| 244 | |
| 245 | if pok { |
| 246 | p.(Pausable).Pause() |
| 247 | } |
| 248 | if gok { |
| 249 | g.Pause() |
| 250 | } |
| 251 | } |
| 252 | |
| 253 | // MendPeer recovers the message dropping behavior of the given peer. |
| 254 | func (t *Transport) MendPeer(id types.ID) { |
| 255 | t.mu.RLock() |
| 256 | p, pok := t.peers[id] |
| 257 | g, gok := t.remotes[id] |
| 258 | t.mu.RUnlock() |
| 259 | |
| 260 | if pok { |
| 261 | p.(Pausable).Resume() |
| 262 | } |
| 263 | if gok { |
| 264 | g.Resume() |
| 265 | } |
| 266 | } |
| 267 | |
| 268 | func (t *Transport) AddRemote(id types.ID, us []string) { |
| 269 | t.mu.Lock() |
| 270 | defer t.mu.Unlock() |
| 271 | if t.remotes == nil { |
| 272 | // there's no clean way to shutdown the golang http server |
| 273 | // (see: https://github.com/golang/go/issues/4674) before |
| 274 | // stopping the transport; ignore any new connections. |
| 275 | return |
| 276 | } |
| 277 | if _, ok := t.peers[id]; ok { |
| 278 | return |
| 279 | } |
| 280 | if _, ok := t.remotes[id]; ok { |
| 281 | return |
| 282 | } |
| 283 | urls, err := types.NewURLs(us) |
| 284 | if err != nil { |
| 285 | if t.Logger != nil { |
| 286 | t.Logger.Panic("failed NewURLs", zap.Strings("urls", us), zap.Error(err)) |
| 287 | } else { |
| 288 | plog.Panicf("newURLs %+v should never fail: %+v", us, err) |
| 289 | } |
| 290 | } |
| 291 | t.remotes[id] = startRemote(t, urls, id) |
| 292 | |
| 293 | if t.Logger != nil { |
| 294 | t.Logger.Info( |
| 295 | "added new remote peer", |
| 296 | zap.String("local-member-id", t.ID.String()), |
| 297 | zap.String("remote-peer-id", id.String()), |
| 298 | zap.Strings("remote-peer-urls", us), |
| 299 | ) |
| 300 | } |
| 301 | } |
| 302 | |
| 303 | func (t *Transport) AddPeer(id types.ID, us []string) { |
| 304 | t.mu.Lock() |
| 305 | defer t.mu.Unlock() |
| 306 | |
| 307 | if t.peers == nil { |
| 308 | panic("transport stopped") |
| 309 | } |
| 310 | if _, ok := t.peers[id]; ok { |
| 311 | return |
| 312 | } |
| 313 | urls, err := types.NewURLs(us) |
| 314 | if err != nil { |
| 315 | if t.Logger != nil { |
| 316 | t.Logger.Panic("failed NewURLs", zap.Strings("urls", us), zap.Error(err)) |
| 317 | } else { |
| 318 | plog.Panicf("newURLs %+v should never fail: %+v", us, err) |
| 319 | } |
| 320 | } |
| 321 | fs := t.LeaderStats.Follower(id.String()) |
| 322 | t.peers[id] = startPeer(t, urls, id, fs) |
| 323 | addPeerToProber(t.Logger, t.pipelineProber, id.String(), us, RoundTripperNameSnapshot, rttSec) |
| 324 | addPeerToProber(t.Logger, t.streamProber, id.String(), us, RoundTripperNameRaftMessage, rttSec) |
| 325 | |
| 326 | if t.Logger != nil { |
| 327 | t.Logger.Info( |
| 328 | "added remote peer", |
| 329 | zap.String("local-member-id", t.ID.String()), |
| 330 | zap.String("remote-peer-id", id.String()), |
| 331 | zap.Strings("remote-peer-urls", us), |
| 332 | ) |
| 333 | } else { |
| 334 | plog.Infof("added peer %s", id) |
| 335 | } |
| 336 | } |
| 337 | |
| 338 | func (t *Transport) RemovePeer(id types.ID) { |
| 339 | t.mu.Lock() |
| 340 | defer t.mu.Unlock() |
| 341 | t.removePeer(id) |
| 342 | } |
| 343 | |
| 344 | func (t *Transport) RemoveAllPeers() { |
| 345 | t.mu.Lock() |
| 346 | defer t.mu.Unlock() |
| 347 | for id := range t.peers { |
| 348 | t.removePeer(id) |
| 349 | } |
| 350 | } |
| 351 | |
| 352 | // the caller of this function must have the peers mutex. |
| 353 | func (t *Transport) removePeer(id types.ID) { |
| 354 | if peer, ok := t.peers[id]; ok { |
| 355 | peer.stop() |
| 356 | } else { |
| 357 | if t.Logger != nil { |
| 358 | t.Logger.Panic("unexpected removal of unknown remote peer", zap.String("remote-peer-id", id.String())) |
| 359 | } else { |
| 360 | plog.Panicf("unexpected removal of unknown peer '%d'", id) |
| 361 | } |
| 362 | } |
| 363 | delete(t.peers, id) |
| 364 | delete(t.LeaderStats.Followers, id.String()) |
| 365 | t.pipelineProber.Remove(id.String()) |
| 366 | t.streamProber.Remove(id.String()) |
| 367 | |
| 368 | if t.Logger != nil { |
| 369 | t.Logger.Info( |
| 370 | "removed remote peer", |
| 371 | zap.String("local-member-id", t.ID.String()), |
| 372 | zap.String("removed-remote-peer-id", id.String()), |
| 373 | ) |
| 374 | } else { |
| 375 | plog.Infof("removed peer %s", id) |
| 376 | } |
| 377 | } |
| 378 | |
| 379 | func (t *Transport) UpdatePeer(id types.ID, us []string) { |
| 380 | t.mu.Lock() |
| 381 | defer t.mu.Unlock() |
| 382 | // TODO: return error or just panic? |
| 383 | if _, ok := t.peers[id]; !ok { |
| 384 | return |
| 385 | } |
| 386 | urls, err := types.NewURLs(us) |
| 387 | if err != nil { |
| 388 | if t.Logger != nil { |
| 389 | t.Logger.Panic("failed NewURLs", zap.Strings("urls", us), zap.Error(err)) |
| 390 | } else { |
| 391 | plog.Panicf("newURLs %+v should never fail: %+v", us, err) |
| 392 | } |
| 393 | } |
| 394 | t.peers[id].update(urls) |
| 395 | |
| 396 | t.pipelineProber.Remove(id.String()) |
| 397 | addPeerToProber(t.Logger, t.pipelineProber, id.String(), us, RoundTripperNameSnapshot, rttSec) |
| 398 | t.streamProber.Remove(id.String()) |
| 399 | addPeerToProber(t.Logger, t.streamProber, id.String(), us, RoundTripperNameRaftMessage, rttSec) |
| 400 | |
| 401 | if t.Logger != nil { |
| 402 | t.Logger.Info( |
| 403 | "updated remote peer", |
| 404 | zap.String("local-member-id", t.ID.String()), |
| 405 | zap.String("updated-remote-peer-id", id.String()), |
| 406 | zap.Strings("updated-remote-peer-urls", us), |
| 407 | ) |
| 408 | } else { |
| 409 | plog.Infof("updated peer %s", id) |
| 410 | } |
| 411 | } |
| 412 | |
| 413 | func (t *Transport) ActiveSince(id types.ID) time.Time { |
| 414 | t.mu.RLock() |
| 415 | defer t.mu.RUnlock() |
| 416 | if p, ok := t.peers[id]; ok { |
| 417 | return p.activeSince() |
| 418 | } |
| 419 | return time.Time{} |
| 420 | } |
| 421 | |
| 422 | func (t *Transport) SendSnapshot(m snap.Message) { |
| 423 | t.mu.Lock() |
| 424 | defer t.mu.Unlock() |
| 425 | p := t.peers[types.ID(m.To)] |
| 426 | if p == nil { |
| 427 | m.CloseWithError(errMemberNotFound) |
| 428 | return |
| 429 | } |
| 430 | p.sendSnap(m) |
| 431 | } |
| 432 | |
| 433 | // Pausable is a testing interface for pausing transport traffic. |
| 434 | type Pausable interface { |
| 435 | Pause() |
| 436 | Resume() |
| 437 | } |
| 438 | |
| 439 | func (t *Transport) Pause() { |
| 440 | t.mu.RLock() |
| 441 | defer t.mu.RUnlock() |
| 442 | for _, p := range t.peers { |
| 443 | p.(Pausable).Pause() |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | func (t *Transport) Resume() { |
| 448 | t.mu.RLock() |
| 449 | defer t.mu.RUnlock() |
| 450 | for _, p := range t.peers { |
| 451 | p.(Pausable).Resume() |
| 452 | } |
| 453 | } |
| 454 | |
| 455 | // ActivePeers returns a channel that closes when an initial |
| 456 | // peer connection has been established. Use this to wait until the |
| 457 | // first peer connection becomes active. |
| 458 | func (t *Transport) ActivePeers() (cnt int) { |
| 459 | t.mu.RLock() |
| 460 | defer t.mu.RUnlock() |
| 461 | for _, p := range t.peers { |
| 462 | if !p.activeSince().IsZero() { |
| 463 | cnt++ |
| 464 | } |
| 465 | } |
| 466 | return cnt |
| 467 | } |