Takahiro Suzuki | d7bf820 | 2020-12-17 20:21:59 +0900 | [diff] [blame] | 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 | |
| 15 | package clientv3 |
| 16 | |
| 17 | import ( |
| 18 | "context" |
| 19 | "sync" |
| 20 | "time" |
| 21 | |
| 22 | "go.etcd.io/etcd/etcdserver/api/v3rpc/rpctypes" |
| 23 | pb "go.etcd.io/etcd/etcdserver/etcdserverpb" |
| 24 | |
| 25 | "go.uber.org/zap" |
| 26 | "google.golang.org/grpc" |
| 27 | "google.golang.org/grpc/metadata" |
| 28 | ) |
| 29 | |
| 30 | type ( |
| 31 | LeaseRevokeResponse pb.LeaseRevokeResponse |
| 32 | LeaseID int64 |
| 33 | ) |
| 34 | |
| 35 | // LeaseGrantResponse wraps the protobuf message LeaseGrantResponse. |
| 36 | type LeaseGrantResponse struct { |
| 37 | *pb.ResponseHeader |
| 38 | ID LeaseID |
| 39 | TTL int64 |
| 40 | Error string |
| 41 | } |
| 42 | |
| 43 | // LeaseKeepAliveResponse wraps the protobuf message LeaseKeepAliveResponse. |
| 44 | type LeaseKeepAliveResponse struct { |
| 45 | *pb.ResponseHeader |
| 46 | ID LeaseID |
| 47 | TTL int64 |
| 48 | } |
| 49 | |
| 50 | // LeaseTimeToLiveResponse wraps the protobuf message LeaseTimeToLiveResponse. |
| 51 | type LeaseTimeToLiveResponse struct { |
| 52 | *pb.ResponseHeader |
| 53 | ID LeaseID `json:"id"` |
| 54 | |
| 55 | // TTL is the remaining TTL in seconds for the lease; the lease will expire in under TTL+1 seconds. Expired lease will return -1. |
| 56 | TTL int64 `json:"ttl"` |
| 57 | |
| 58 | // GrantedTTL is the initial granted time in seconds upon lease creation/renewal. |
| 59 | GrantedTTL int64 `json:"granted-ttl"` |
| 60 | |
| 61 | // Keys is the list of keys attached to this lease. |
| 62 | Keys [][]byte `json:"keys"` |
| 63 | } |
| 64 | |
| 65 | // LeaseStatus represents a lease status. |
| 66 | type LeaseStatus struct { |
| 67 | ID LeaseID `json:"id"` |
| 68 | // TODO: TTL int64 |
| 69 | } |
| 70 | |
| 71 | // LeaseLeasesResponse wraps the protobuf message LeaseLeasesResponse. |
| 72 | type LeaseLeasesResponse struct { |
| 73 | *pb.ResponseHeader |
| 74 | Leases []LeaseStatus `json:"leases"` |
| 75 | } |
| 76 | |
| 77 | const ( |
| 78 | // defaultTTL is the assumed lease TTL used for the first keepalive |
| 79 | // deadline before the actual TTL is known to the client. |
| 80 | defaultTTL = 5 * time.Second |
| 81 | // NoLease is a lease ID for the absence of a lease. |
| 82 | NoLease LeaseID = 0 |
| 83 | |
| 84 | // retryConnWait is how long to wait before retrying request due to an error |
| 85 | retryConnWait = 500 * time.Millisecond |
| 86 | ) |
| 87 | |
| 88 | // LeaseResponseChSize is the size of buffer to store unsent lease responses. |
| 89 | // WARNING: DO NOT UPDATE. |
| 90 | // Only for testing purposes. |
| 91 | var LeaseResponseChSize = 16 |
| 92 | |
| 93 | // ErrKeepAliveHalted is returned if client keep alive loop halts with an unexpected error. |
| 94 | // |
| 95 | // This usually means that automatic lease renewal via KeepAlive is broken, but KeepAliveOnce will still work as expected. |
| 96 | type ErrKeepAliveHalted struct { |
| 97 | Reason error |
| 98 | } |
| 99 | |
| 100 | func (e ErrKeepAliveHalted) Error() string { |
| 101 | s := "etcdclient: leases keep alive halted" |
| 102 | if e.Reason != nil { |
| 103 | s += ": " + e.Reason.Error() |
| 104 | } |
| 105 | return s |
| 106 | } |
| 107 | |
| 108 | type Lease interface { |
| 109 | // Grant creates a new lease. |
| 110 | Grant(ctx context.Context, ttl int64) (*LeaseGrantResponse, error) |
| 111 | |
| 112 | // Revoke revokes the given lease. |
| 113 | Revoke(ctx context.Context, id LeaseID) (*LeaseRevokeResponse, error) |
| 114 | |
| 115 | // TimeToLive retrieves the lease information of the given lease ID. |
| 116 | TimeToLive(ctx context.Context, id LeaseID, opts ...LeaseOption) (*LeaseTimeToLiveResponse, error) |
| 117 | |
| 118 | // Leases retrieves all leases. |
| 119 | Leases(ctx context.Context) (*LeaseLeasesResponse, error) |
| 120 | |
| 121 | // KeepAlive attempts to keep the given lease alive forever. If the keepalive responses posted |
| 122 | // to the channel are not consumed promptly the channel may become full. When full, the lease |
| 123 | // client will continue sending keep alive requests to the etcd server, but will drop responses |
| 124 | // until there is capacity on the channel to send more responses. |
| 125 | // |
| 126 | // If client keep alive loop halts with an unexpected error (e.g. "etcdserver: no leader") or |
| 127 | // canceled by the caller (e.g. context.Canceled), KeepAlive returns a ErrKeepAliveHalted error |
| 128 | // containing the error reason. |
| 129 | // |
| 130 | // The returned "LeaseKeepAliveResponse" channel closes if underlying keep |
| 131 | // alive stream is interrupted in some way the client cannot handle itself; |
| 132 | // given context "ctx" is canceled or timed out. |
| 133 | // |
| 134 | // TODO(v4.0): post errors to last keep alive message before closing |
| 135 | // (see https://github.com/etcd-io/etcd/pull/7866) |
| 136 | KeepAlive(ctx context.Context, id LeaseID) (<-chan *LeaseKeepAliveResponse, error) |
| 137 | |
| 138 | // KeepAliveOnce renews the lease once. The response corresponds to the |
| 139 | // first message from calling KeepAlive. If the response has a recoverable |
| 140 | // error, KeepAliveOnce will retry the RPC with a new keep alive message. |
| 141 | // |
| 142 | // In most of the cases, Keepalive should be used instead of KeepAliveOnce. |
| 143 | KeepAliveOnce(ctx context.Context, id LeaseID) (*LeaseKeepAliveResponse, error) |
| 144 | |
| 145 | // Close releases all resources Lease keeps for efficient communication |
| 146 | // with the etcd server. |
| 147 | Close() error |
| 148 | } |
| 149 | |
| 150 | type lessor struct { |
| 151 | mu sync.Mutex // guards all fields |
| 152 | |
| 153 | // donec is closed and loopErr is set when recvKeepAliveLoop stops |
| 154 | donec chan struct{} |
| 155 | loopErr error |
| 156 | |
| 157 | remote pb.LeaseClient |
| 158 | |
| 159 | stream pb.Lease_LeaseKeepAliveClient |
| 160 | streamCancel context.CancelFunc |
| 161 | |
| 162 | stopCtx context.Context |
| 163 | stopCancel context.CancelFunc |
| 164 | |
| 165 | keepAlives map[LeaseID]*keepAlive |
| 166 | |
| 167 | // firstKeepAliveTimeout is the timeout for the first keepalive request |
| 168 | // before the actual TTL is known to the lease client |
| 169 | firstKeepAliveTimeout time.Duration |
| 170 | |
| 171 | // firstKeepAliveOnce ensures stream starts after first KeepAlive call. |
| 172 | firstKeepAliveOnce sync.Once |
| 173 | |
| 174 | callOpts []grpc.CallOption |
| 175 | |
| 176 | lg *zap.Logger |
| 177 | } |
| 178 | |
| 179 | // keepAlive multiplexes a keepalive for a lease over multiple channels |
| 180 | type keepAlive struct { |
| 181 | chs []chan<- *LeaseKeepAliveResponse |
| 182 | ctxs []context.Context |
| 183 | // deadline is the time the keep alive channels close if no response |
| 184 | deadline time.Time |
| 185 | // nextKeepAlive is when to send the next keep alive message |
| 186 | nextKeepAlive time.Time |
| 187 | // donec is closed on lease revoke, expiration, or cancel. |
| 188 | donec chan struct{} |
| 189 | } |
| 190 | |
| 191 | func NewLease(c *Client) Lease { |
| 192 | return NewLeaseFromLeaseClient(RetryLeaseClient(c), c, c.cfg.DialTimeout+time.Second) |
| 193 | } |
| 194 | |
| 195 | func NewLeaseFromLeaseClient(remote pb.LeaseClient, c *Client, keepAliveTimeout time.Duration) Lease { |
| 196 | l := &lessor{ |
| 197 | donec: make(chan struct{}), |
| 198 | keepAlives: make(map[LeaseID]*keepAlive), |
| 199 | remote: remote, |
| 200 | firstKeepAliveTimeout: keepAliveTimeout, |
| 201 | lg: c.lg, |
| 202 | } |
| 203 | if l.firstKeepAliveTimeout == time.Second { |
| 204 | l.firstKeepAliveTimeout = defaultTTL |
| 205 | } |
| 206 | if c != nil { |
| 207 | l.callOpts = c.callOpts |
| 208 | } |
| 209 | reqLeaderCtx := WithRequireLeader(context.Background()) |
| 210 | l.stopCtx, l.stopCancel = context.WithCancel(reqLeaderCtx) |
| 211 | return l |
| 212 | } |
| 213 | |
| 214 | func (l *lessor) Grant(ctx context.Context, ttl int64) (*LeaseGrantResponse, error) { |
| 215 | r := &pb.LeaseGrantRequest{TTL: ttl} |
| 216 | resp, err := l.remote.LeaseGrant(ctx, r, l.callOpts...) |
| 217 | if err == nil { |
| 218 | gresp := &LeaseGrantResponse{ |
| 219 | ResponseHeader: resp.GetHeader(), |
| 220 | ID: LeaseID(resp.ID), |
| 221 | TTL: resp.TTL, |
| 222 | Error: resp.Error, |
| 223 | } |
| 224 | return gresp, nil |
| 225 | } |
| 226 | return nil, toErr(ctx, err) |
| 227 | } |
| 228 | |
| 229 | func (l *lessor) Revoke(ctx context.Context, id LeaseID) (*LeaseRevokeResponse, error) { |
| 230 | r := &pb.LeaseRevokeRequest{ID: int64(id)} |
| 231 | resp, err := l.remote.LeaseRevoke(ctx, r, l.callOpts...) |
| 232 | if err == nil { |
| 233 | return (*LeaseRevokeResponse)(resp), nil |
| 234 | } |
| 235 | return nil, toErr(ctx, err) |
| 236 | } |
| 237 | |
| 238 | func (l *lessor) TimeToLive(ctx context.Context, id LeaseID, opts ...LeaseOption) (*LeaseTimeToLiveResponse, error) { |
| 239 | r := toLeaseTimeToLiveRequest(id, opts...) |
| 240 | resp, err := l.remote.LeaseTimeToLive(ctx, r, l.callOpts...) |
| 241 | if err == nil { |
| 242 | gresp := &LeaseTimeToLiveResponse{ |
| 243 | ResponseHeader: resp.GetHeader(), |
| 244 | ID: LeaseID(resp.ID), |
| 245 | TTL: resp.TTL, |
| 246 | GrantedTTL: resp.GrantedTTL, |
| 247 | Keys: resp.Keys, |
| 248 | } |
| 249 | return gresp, nil |
| 250 | } |
| 251 | return nil, toErr(ctx, err) |
| 252 | } |
| 253 | |
| 254 | func (l *lessor) Leases(ctx context.Context) (*LeaseLeasesResponse, error) { |
| 255 | resp, err := l.remote.LeaseLeases(ctx, &pb.LeaseLeasesRequest{}, l.callOpts...) |
| 256 | if err == nil { |
| 257 | leases := make([]LeaseStatus, len(resp.Leases)) |
| 258 | for i := range resp.Leases { |
| 259 | leases[i] = LeaseStatus{ID: LeaseID(resp.Leases[i].ID)} |
| 260 | } |
| 261 | return &LeaseLeasesResponse{ResponseHeader: resp.GetHeader(), Leases: leases}, nil |
| 262 | } |
| 263 | return nil, toErr(ctx, err) |
| 264 | } |
| 265 | |
| 266 | func (l *lessor) KeepAlive(ctx context.Context, id LeaseID) (<-chan *LeaseKeepAliveResponse, error) { |
| 267 | ch := make(chan *LeaseKeepAliveResponse, LeaseResponseChSize) |
| 268 | |
| 269 | l.mu.Lock() |
| 270 | // ensure that recvKeepAliveLoop is still running |
| 271 | select { |
| 272 | case <-l.donec: |
| 273 | err := l.loopErr |
| 274 | l.mu.Unlock() |
| 275 | close(ch) |
| 276 | return ch, ErrKeepAliveHalted{Reason: err} |
| 277 | default: |
| 278 | } |
| 279 | ka, ok := l.keepAlives[id] |
| 280 | if !ok { |
| 281 | // create fresh keep alive |
| 282 | ka = &keepAlive{ |
| 283 | chs: []chan<- *LeaseKeepAliveResponse{ch}, |
| 284 | ctxs: []context.Context{ctx}, |
| 285 | deadline: time.Now().Add(l.firstKeepAliveTimeout), |
| 286 | nextKeepAlive: time.Now(), |
| 287 | donec: make(chan struct{}), |
| 288 | } |
| 289 | l.keepAlives[id] = ka |
| 290 | } else { |
| 291 | // add channel and context to existing keep alive |
| 292 | ka.ctxs = append(ka.ctxs, ctx) |
| 293 | ka.chs = append(ka.chs, ch) |
| 294 | } |
| 295 | l.mu.Unlock() |
| 296 | |
| 297 | go l.keepAliveCtxCloser(ctx, id, ka.donec) |
| 298 | l.firstKeepAliveOnce.Do(func() { |
| 299 | go l.recvKeepAliveLoop() |
| 300 | go l.deadlineLoop() |
| 301 | }) |
| 302 | |
| 303 | return ch, nil |
| 304 | } |
| 305 | |
| 306 | func (l *lessor) KeepAliveOnce(ctx context.Context, id LeaseID) (*LeaseKeepAliveResponse, error) { |
| 307 | for { |
| 308 | resp, err := l.keepAliveOnce(ctx, id) |
| 309 | if err == nil { |
| 310 | if resp.TTL <= 0 { |
| 311 | err = rpctypes.ErrLeaseNotFound |
| 312 | } |
| 313 | return resp, err |
| 314 | } |
| 315 | if isHaltErr(ctx, err) { |
| 316 | return nil, toErr(ctx, err) |
| 317 | } |
| 318 | } |
| 319 | } |
| 320 | |
| 321 | func (l *lessor) Close() error { |
| 322 | l.stopCancel() |
| 323 | // close for synchronous teardown if stream goroutines never launched |
| 324 | l.firstKeepAliveOnce.Do(func() { close(l.donec) }) |
| 325 | <-l.donec |
| 326 | return nil |
| 327 | } |
| 328 | |
| 329 | func (l *lessor) keepAliveCtxCloser(ctx context.Context, id LeaseID, donec <-chan struct{}) { |
| 330 | select { |
| 331 | case <-donec: |
| 332 | return |
| 333 | case <-l.donec: |
| 334 | return |
| 335 | case <-ctx.Done(): |
| 336 | } |
| 337 | |
| 338 | l.mu.Lock() |
| 339 | defer l.mu.Unlock() |
| 340 | |
| 341 | ka, ok := l.keepAlives[id] |
| 342 | if !ok { |
| 343 | return |
| 344 | } |
| 345 | |
| 346 | // close channel and remove context if still associated with keep alive |
| 347 | for i, c := range ka.ctxs { |
| 348 | if c == ctx { |
| 349 | close(ka.chs[i]) |
| 350 | ka.ctxs = append(ka.ctxs[:i], ka.ctxs[i+1:]...) |
| 351 | ka.chs = append(ka.chs[:i], ka.chs[i+1:]...) |
| 352 | break |
| 353 | } |
| 354 | } |
| 355 | // remove if no one more listeners |
| 356 | if len(ka.chs) == 0 { |
| 357 | delete(l.keepAlives, id) |
| 358 | } |
| 359 | } |
| 360 | |
| 361 | // closeRequireLeader scans keepAlives for ctxs that have require leader |
| 362 | // and closes the associated channels. |
| 363 | func (l *lessor) closeRequireLeader() { |
| 364 | l.mu.Lock() |
| 365 | defer l.mu.Unlock() |
| 366 | for _, ka := range l.keepAlives { |
| 367 | reqIdxs := 0 |
| 368 | // find all required leader channels, close, mark as nil |
| 369 | for i, ctx := range ka.ctxs { |
| 370 | md, ok := metadata.FromOutgoingContext(ctx) |
| 371 | if !ok { |
| 372 | continue |
| 373 | } |
| 374 | ks := md[rpctypes.MetadataRequireLeaderKey] |
| 375 | if len(ks) < 1 || ks[0] != rpctypes.MetadataHasLeader { |
| 376 | continue |
| 377 | } |
| 378 | close(ka.chs[i]) |
| 379 | ka.chs[i] = nil |
| 380 | reqIdxs++ |
| 381 | } |
| 382 | if reqIdxs == 0 { |
| 383 | continue |
| 384 | } |
| 385 | // remove all channels that required a leader from keepalive |
| 386 | newChs := make([]chan<- *LeaseKeepAliveResponse, len(ka.chs)-reqIdxs) |
| 387 | newCtxs := make([]context.Context, len(newChs)) |
| 388 | newIdx := 0 |
| 389 | for i := range ka.chs { |
| 390 | if ka.chs[i] == nil { |
| 391 | continue |
| 392 | } |
| 393 | newChs[newIdx], newCtxs[newIdx] = ka.chs[i], ka.ctxs[newIdx] |
| 394 | newIdx++ |
| 395 | } |
| 396 | ka.chs, ka.ctxs = newChs, newCtxs |
| 397 | } |
| 398 | } |
| 399 | |
| 400 | func (l *lessor) keepAliveOnce(ctx context.Context, id LeaseID) (*LeaseKeepAliveResponse, error) { |
| 401 | cctx, cancel := context.WithCancel(ctx) |
| 402 | defer cancel() |
| 403 | |
| 404 | stream, err := l.remote.LeaseKeepAlive(cctx, l.callOpts...) |
| 405 | if err != nil { |
| 406 | return nil, toErr(ctx, err) |
| 407 | } |
| 408 | |
| 409 | err = stream.Send(&pb.LeaseKeepAliveRequest{ID: int64(id)}) |
| 410 | if err != nil { |
| 411 | return nil, toErr(ctx, err) |
| 412 | } |
| 413 | |
| 414 | resp, rerr := stream.Recv() |
| 415 | if rerr != nil { |
| 416 | return nil, toErr(ctx, rerr) |
| 417 | } |
| 418 | |
| 419 | karesp := &LeaseKeepAliveResponse{ |
| 420 | ResponseHeader: resp.GetHeader(), |
| 421 | ID: LeaseID(resp.ID), |
| 422 | TTL: resp.TTL, |
| 423 | } |
| 424 | return karesp, nil |
| 425 | } |
| 426 | |
| 427 | func (l *lessor) recvKeepAliveLoop() (gerr error) { |
| 428 | defer func() { |
| 429 | l.mu.Lock() |
| 430 | close(l.donec) |
| 431 | l.loopErr = gerr |
| 432 | for _, ka := range l.keepAlives { |
| 433 | ka.close() |
| 434 | } |
| 435 | l.keepAlives = make(map[LeaseID]*keepAlive) |
| 436 | l.mu.Unlock() |
| 437 | }() |
| 438 | |
| 439 | for { |
| 440 | stream, err := l.resetRecv() |
| 441 | if err != nil { |
| 442 | if canceledByCaller(l.stopCtx, err) { |
| 443 | return err |
| 444 | } |
| 445 | } else { |
| 446 | for { |
| 447 | resp, err := stream.Recv() |
| 448 | if err != nil { |
| 449 | if canceledByCaller(l.stopCtx, err) { |
| 450 | return err |
| 451 | } |
| 452 | |
| 453 | if toErr(l.stopCtx, err) == rpctypes.ErrNoLeader { |
| 454 | l.closeRequireLeader() |
| 455 | } |
| 456 | break |
| 457 | } |
| 458 | |
| 459 | l.recvKeepAlive(resp) |
| 460 | } |
| 461 | } |
| 462 | |
| 463 | select { |
| 464 | case <-time.After(retryConnWait): |
| 465 | case <-l.stopCtx.Done(): |
| 466 | return l.stopCtx.Err() |
| 467 | } |
| 468 | } |
| 469 | } |
| 470 | |
| 471 | // resetRecv opens a new lease stream and starts sending keep alive requests. |
| 472 | func (l *lessor) resetRecv() (pb.Lease_LeaseKeepAliveClient, error) { |
| 473 | sctx, cancel := context.WithCancel(l.stopCtx) |
| 474 | stream, err := l.remote.LeaseKeepAlive(sctx, append(l.callOpts, withMax(0))...) |
| 475 | if err != nil { |
| 476 | cancel() |
| 477 | return nil, err |
| 478 | } |
| 479 | |
| 480 | l.mu.Lock() |
| 481 | defer l.mu.Unlock() |
| 482 | if l.stream != nil && l.streamCancel != nil { |
| 483 | l.streamCancel() |
| 484 | } |
| 485 | |
| 486 | l.streamCancel = cancel |
| 487 | l.stream = stream |
| 488 | |
| 489 | go l.sendKeepAliveLoop(stream) |
| 490 | return stream, nil |
| 491 | } |
| 492 | |
| 493 | // recvKeepAlive updates a lease based on its LeaseKeepAliveResponse |
| 494 | func (l *lessor) recvKeepAlive(resp *pb.LeaseKeepAliveResponse) { |
| 495 | karesp := &LeaseKeepAliveResponse{ |
| 496 | ResponseHeader: resp.GetHeader(), |
| 497 | ID: LeaseID(resp.ID), |
| 498 | TTL: resp.TTL, |
| 499 | } |
| 500 | |
| 501 | l.mu.Lock() |
| 502 | defer l.mu.Unlock() |
| 503 | |
| 504 | ka, ok := l.keepAlives[karesp.ID] |
| 505 | if !ok { |
| 506 | return |
| 507 | } |
| 508 | |
| 509 | if karesp.TTL <= 0 { |
| 510 | // lease expired; close all keep alive channels |
| 511 | delete(l.keepAlives, karesp.ID) |
| 512 | ka.close() |
| 513 | return |
| 514 | } |
| 515 | |
| 516 | // send update to all channels |
| 517 | nextKeepAlive := time.Now().Add((time.Duration(karesp.TTL) * time.Second) / 3.0) |
| 518 | ka.deadline = time.Now().Add(time.Duration(karesp.TTL) * time.Second) |
| 519 | for _, ch := range ka.chs { |
| 520 | select { |
| 521 | case ch <- karesp: |
| 522 | default: |
| 523 | if l.lg != nil { |
| 524 | l.lg.Warn("lease keepalive response queue is full; dropping response send", |
| 525 | zap.Int("queue-size", len(ch)), |
| 526 | zap.Int("queue-capacity", cap(ch)), |
| 527 | ) |
| 528 | } |
| 529 | } |
| 530 | // still advance in order to rate-limit keep-alive sends |
| 531 | ka.nextKeepAlive = nextKeepAlive |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | // deadlineLoop reaps any keep alive channels that have not received a response |
| 536 | // within the lease TTL |
| 537 | func (l *lessor) deadlineLoop() { |
| 538 | for { |
| 539 | select { |
| 540 | case <-time.After(time.Second): |
| 541 | case <-l.donec: |
| 542 | return |
| 543 | } |
| 544 | now := time.Now() |
| 545 | l.mu.Lock() |
| 546 | for id, ka := range l.keepAlives { |
| 547 | if ka.deadline.Before(now) { |
| 548 | // waited too long for response; lease may be expired |
| 549 | ka.close() |
| 550 | delete(l.keepAlives, id) |
| 551 | } |
| 552 | } |
| 553 | l.mu.Unlock() |
| 554 | } |
| 555 | } |
| 556 | |
| 557 | // sendKeepAliveLoop sends keep alive requests for the lifetime of the given stream. |
| 558 | func (l *lessor) sendKeepAliveLoop(stream pb.Lease_LeaseKeepAliveClient) { |
| 559 | for { |
| 560 | var tosend []LeaseID |
| 561 | |
| 562 | now := time.Now() |
| 563 | l.mu.Lock() |
| 564 | for id, ka := range l.keepAlives { |
| 565 | if ka.nextKeepAlive.Before(now) { |
| 566 | tosend = append(tosend, id) |
| 567 | } |
| 568 | } |
| 569 | l.mu.Unlock() |
| 570 | |
| 571 | for _, id := range tosend { |
| 572 | r := &pb.LeaseKeepAliveRequest{ID: int64(id)} |
| 573 | if err := stream.Send(r); err != nil { |
| 574 | // TODO do something with this error? |
| 575 | return |
| 576 | } |
| 577 | } |
| 578 | |
| 579 | select { |
| 580 | case <-time.After(retryConnWait): |
| 581 | case <-stream.Context().Done(): |
| 582 | return |
| 583 | case <-l.donec: |
| 584 | return |
| 585 | case <-l.stopCtx.Done(): |
| 586 | return |
| 587 | } |
| 588 | } |
| 589 | } |
| 590 | |
| 591 | func (ka *keepAlive) close() { |
| 592 | close(ka.donec) |
| 593 | for _, ch := range ka.chs { |
| 594 | close(ch) |
| 595 | } |
| 596 | } |