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 wal |
| 16 | |
| 17 | import ( |
| 18 | "bytes" |
| 19 | "errors" |
| 20 | "fmt" |
| 21 | "hash/crc32" |
| 22 | "io" |
| 23 | "os" |
| 24 | "path/filepath" |
| 25 | "sync" |
| 26 | "time" |
| 27 | |
| 28 | "go.etcd.io/etcd/pkg/fileutil" |
| 29 | "go.etcd.io/etcd/pkg/pbutil" |
| 30 | "go.etcd.io/etcd/raft" |
| 31 | "go.etcd.io/etcd/raft/raftpb" |
| 32 | "go.etcd.io/etcd/wal/walpb" |
| 33 | |
| 34 | "github.com/coreos/pkg/capnslog" |
| 35 | "go.uber.org/zap" |
| 36 | ) |
| 37 | |
| 38 | const ( |
| 39 | metadataType int64 = iota + 1 |
| 40 | entryType |
| 41 | stateType |
| 42 | crcType |
| 43 | snapshotType |
| 44 | |
| 45 | // warnSyncDuration is the amount of time allotted to an fsync before |
| 46 | // logging a warning |
| 47 | warnSyncDuration = time.Second |
| 48 | ) |
| 49 | |
| 50 | var ( |
| 51 | // SegmentSizeBytes is the preallocated size of each wal segment file. |
| 52 | // The actual size might be larger than this. In general, the default |
| 53 | // value should be used, but this is defined as an exported variable |
| 54 | // so that tests can set a different segment size. |
| 55 | SegmentSizeBytes int64 = 64 * 1000 * 1000 // 64MB |
| 56 | |
| 57 | plog = capnslog.NewPackageLogger("go.etcd.io/etcd", "wal") |
| 58 | |
| 59 | ErrMetadataConflict = errors.New("wal: conflicting metadata found") |
| 60 | ErrFileNotFound = errors.New("wal: file not found") |
| 61 | ErrCRCMismatch = errors.New("wal: crc mismatch") |
| 62 | ErrSnapshotMismatch = errors.New("wal: snapshot mismatch") |
| 63 | ErrSnapshotNotFound = errors.New("wal: snapshot not found") |
| 64 | crcTable = crc32.MakeTable(crc32.Castagnoli) |
| 65 | ) |
| 66 | |
| 67 | // WAL is a logical representation of the stable storage. |
| 68 | // WAL is either in read mode or append mode but not both. |
| 69 | // A newly created WAL is in append mode, and ready for appending records. |
| 70 | // A just opened WAL is in read mode, and ready for reading records. |
| 71 | // The WAL will be ready for appending after reading out all the previous records. |
| 72 | type WAL struct { |
| 73 | lg *zap.Logger |
| 74 | |
| 75 | dir string // the living directory of the underlay files |
| 76 | |
| 77 | // dirFile is a fd for the wal directory for syncing on Rename |
| 78 | dirFile *os.File |
| 79 | |
| 80 | metadata []byte // metadata recorded at the head of each WAL |
| 81 | state raftpb.HardState // hardstate recorded at the head of WAL |
| 82 | |
| 83 | start walpb.Snapshot // snapshot to start reading |
| 84 | decoder *decoder // decoder to decode records |
| 85 | readClose func() error // closer for decode reader |
| 86 | |
| 87 | mu sync.Mutex |
| 88 | enti uint64 // index of the last entry saved to the wal |
| 89 | encoder *encoder // encoder to encode records |
| 90 | |
| 91 | locks []*fileutil.LockedFile // the locked files the WAL holds (the name is increasing) |
| 92 | fp *filePipeline |
| 93 | } |
| 94 | |
| 95 | // Create creates a WAL ready for appending records. The given metadata is |
| 96 | // recorded at the head of each WAL file, and can be retrieved with ReadAll. |
| 97 | func Create(lg *zap.Logger, dirpath string, metadata []byte) (*WAL, error) { |
| 98 | if Exist(dirpath) { |
| 99 | return nil, os.ErrExist |
| 100 | } |
| 101 | |
| 102 | // keep temporary wal directory so WAL initialization appears atomic |
| 103 | tmpdirpath := filepath.Clean(dirpath) + ".tmp" |
| 104 | if fileutil.Exist(tmpdirpath) { |
| 105 | if err := os.RemoveAll(tmpdirpath); err != nil { |
| 106 | return nil, err |
| 107 | } |
| 108 | } |
| 109 | if err := fileutil.CreateDirAll(tmpdirpath); err != nil { |
| 110 | if lg != nil { |
| 111 | lg.Warn( |
| 112 | "failed to create a temporary WAL directory", |
| 113 | zap.String("tmp-dir-path", tmpdirpath), |
| 114 | zap.String("dir-path", dirpath), |
| 115 | zap.Error(err), |
| 116 | ) |
| 117 | } |
| 118 | return nil, err |
| 119 | } |
| 120 | |
| 121 | p := filepath.Join(tmpdirpath, walName(0, 0)) |
| 122 | f, err := fileutil.LockFile(p, os.O_WRONLY|os.O_CREATE, fileutil.PrivateFileMode) |
| 123 | if err != nil { |
| 124 | if lg != nil { |
| 125 | lg.Warn( |
| 126 | "failed to flock an initial WAL file", |
| 127 | zap.String("path", p), |
| 128 | zap.Error(err), |
| 129 | ) |
| 130 | } |
| 131 | return nil, err |
| 132 | } |
| 133 | if _, err = f.Seek(0, io.SeekEnd); err != nil { |
| 134 | if lg != nil { |
| 135 | lg.Warn( |
| 136 | "failed to seek an initial WAL file", |
| 137 | zap.String("path", p), |
| 138 | zap.Error(err), |
| 139 | ) |
| 140 | } |
| 141 | return nil, err |
| 142 | } |
| 143 | if err = fileutil.Preallocate(f.File, SegmentSizeBytes, true); err != nil { |
| 144 | if lg != nil { |
| 145 | lg.Warn( |
| 146 | "failed to preallocate an initial WAL file", |
| 147 | zap.String("path", p), |
| 148 | zap.Int64("segment-bytes", SegmentSizeBytes), |
| 149 | zap.Error(err), |
| 150 | ) |
| 151 | } |
| 152 | return nil, err |
| 153 | } |
| 154 | |
| 155 | w := &WAL{ |
| 156 | lg: lg, |
| 157 | dir: dirpath, |
| 158 | metadata: metadata, |
| 159 | } |
| 160 | w.encoder, err = newFileEncoder(f.File, 0) |
| 161 | if err != nil { |
| 162 | return nil, err |
| 163 | } |
| 164 | w.locks = append(w.locks, f) |
| 165 | if err = w.saveCrc(0); err != nil { |
| 166 | return nil, err |
| 167 | } |
| 168 | if err = w.encoder.encode(&walpb.Record{Type: metadataType, Data: metadata}); err != nil { |
| 169 | return nil, err |
| 170 | } |
| 171 | if err = w.SaveSnapshot(walpb.Snapshot{}); err != nil { |
| 172 | return nil, err |
| 173 | } |
| 174 | |
| 175 | if w, err = w.renameWAL(tmpdirpath); err != nil { |
| 176 | if lg != nil { |
| 177 | lg.Warn( |
| 178 | "failed to rename the temporary WAL directory", |
| 179 | zap.String("tmp-dir-path", tmpdirpath), |
| 180 | zap.String("dir-path", w.dir), |
| 181 | zap.Error(err), |
| 182 | ) |
| 183 | } |
| 184 | return nil, err |
| 185 | } |
| 186 | |
| 187 | var perr error |
| 188 | defer func() { |
| 189 | if perr != nil { |
| 190 | w.cleanupWAL(lg) |
| 191 | } |
| 192 | }() |
| 193 | |
| 194 | // directory was renamed; sync parent dir to persist rename |
| 195 | pdir, perr := fileutil.OpenDir(filepath.Dir(w.dir)) |
| 196 | if perr != nil { |
| 197 | if lg != nil { |
| 198 | lg.Warn( |
| 199 | "failed to open the parent data directory", |
| 200 | zap.String("parent-dir-path", filepath.Dir(w.dir)), |
| 201 | zap.String("dir-path", w.dir), |
| 202 | zap.Error(perr), |
| 203 | ) |
| 204 | } |
| 205 | return nil, perr |
| 206 | } |
| 207 | if perr = fileutil.Fsync(pdir); perr != nil { |
| 208 | if lg != nil { |
| 209 | lg.Warn( |
| 210 | "failed to fsync the parent data directory file", |
| 211 | zap.String("parent-dir-path", filepath.Dir(w.dir)), |
| 212 | zap.String("dir-path", w.dir), |
| 213 | zap.Error(perr), |
| 214 | ) |
| 215 | } |
| 216 | return nil, perr |
| 217 | } |
| 218 | if perr = pdir.Close(); perr != nil { |
| 219 | if lg != nil { |
| 220 | lg.Warn( |
| 221 | "failed to close the parent data directory file", |
| 222 | zap.String("parent-dir-path", filepath.Dir(w.dir)), |
| 223 | zap.String("dir-path", w.dir), |
| 224 | zap.Error(perr), |
| 225 | ) |
| 226 | } |
| 227 | return nil, perr |
| 228 | } |
| 229 | |
| 230 | return w, nil |
| 231 | } |
| 232 | |
| 233 | func (w *WAL) cleanupWAL(lg *zap.Logger) { |
| 234 | var err error |
| 235 | if err = w.Close(); err != nil { |
| 236 | if lg != nil { |
| 237 | lg.Panic("failed to close WAL during cleanup", zap.Error(err)) |
| 238 | } else { |
| 239 | plog.Panicf("failed to close WAL during cleanup: %v", err) |
| 240 | } |
| 241 | } |
| 242 | brokenDirName := fmt.Sprintf("%s.broken.%v", w.dir, time.Now().Format("20060102.150405.999999")) |
| 243 | if err = os.Rename(w.dir, brokenDirName); err != nil { |
| 244 | if lg != nil { |
| 245 | lg.Panic( |
| 246 | "failed to rename WAL during cleanup", |
| 247 | zap.Error(err), |
| 248 | zap.String("source-path", w.dir), |
| 249 | zap.String("rename-path", brokenDirName), |
| 250 | ) |
| 251 | } else { |
| 252 | plog.Panicf("failed to rename WAL during cleanup: %v", err) |
| 253 | } |
| 254 | } |
| 255 | } |
| 256 | |
| 257 | func (w *WAL) renameWAL(tmpdirpath string) (*WAL, error) { |
| 258 | if err := os.RemoveAll(w.dir); err != nil { |
| 259 | return nil, err |
| 260 | } |
| 261 | // On non-Windows platforms, hold the lock while renaming. Releasing |
| 262 | // the lock and trying to reacquire it quickly can be flaky because |
| 263 | // it's possible the process will fork to spawn a process while this is |
| 264 | // happening. The fds are set up as close-on-exec by the Go runtime, |
| 265 | // but there is a window between the fork and the exec where another |
| 266 | // process holds the lock. |
| 267 | if err := os.Rename(tmpdirpath, w.dir); err != nil { |
| 268 | if _, ok := err.(*os.LinkError); ok { |
| 269 | return w.renameWALUnlock(tmpdirpath) |
| 270 | } |
| 271 | return nil, err |
| 272 | } |
| 273 | w.fp = newFilePipeline(w.lg, w.dir, SegmentSizeBytes) |
| 274 | df, err := fileutil.OpenDir(w.dir) |
| 275 | w.dirFile = df |
| 276 | return w, err |
| 277 | } |
| 278 | |
| 279 | func (w *WAL) renameWALUnlock(tmpdirpath string) (*WAL, error) { |
| 280 | // rename of directory with locked files doesn't work on windows/cifs; |
| 281 | // close the WAL to release the locks so the directory can be renamed. |
| 282 | if w.lg != nil { |
| 283 | w.lg.Info( |
| 284 | "closing WAL to release flock and retry directory renaming", |
| 285 | zap.String("from", tmpdirpath), |
| 286 | zap.String("to", w.dir), |
| 287 | ) |
| 288 | } else { |
| 289 | plog.Infof("releasing file lock to rename %q to %q", tmpdirpath, w.dir) |
| 290 | } |
| 291 | w.Close() |
| 292 | |
| 293 | if err := os.Rename(tmpdirpath, w.dir); err != nil { |
| 294 | return nil, err |
| 295 | } |
| 296 | |
| 297 | // reopen and relock |
| 298 | newWAL, oerr := Open(w.lg, w.dir, walpb.Snapshot{}) |
| 299 | if oerr != nil { |
| 300 | return nil, oerr |
| 301 | } |
| 302 | if _, _, _, err := newWAL.ReadAll(); err != nil { |
| 303 | newWAL.Close() |
| 304 | return nil, err |
| 305 | } |
| 306 | return newWAL, nil |
| 307 | } |
| 308 | |
| 309 | // Open opens the WAL at the given snap. |
| 310 | // The snap SHOULD have been previously saved to the WAL, or the following |
| 311 | // ReadAll will fail. |
| 312 | // The returned WAL is ready to read and the first record will be the one after |
| 313 | // the given snap. The WAL cannot be appended to before reading out all of its |
| 314 | // previous records. |
| 315 | func Open(lg *zap.Logger, dirpath string, snap walpb.Snapshot) (*WAL, error) { |
| 316 | w, err := openAtIndex(lg, dirpath, snap, true) |
| 317 | if err != nil { |
| 318 | return nil, err |
| 319 | } |
| 320 | if w.dirFile, err = fileutil.OpenDir(w.dir); err != nil { |
| 321 | return nil, err |
| 322 | } |
| 323 | return w, nil |
| 324 | } |
| 325 | |
| 326 | // OpenForRead only opens the wal files for read. |
| 327 | // Write on a read only wal panics. |
| 328 | func OpenForRead(lg *zap.Logger, dirpath string, snap walpb.Snapshot) (*WAL, error) { |
| 329 | return openAtIndex(lg, dirpath, snap, false) |
| 330 | } |
| 331 | |
| 332 | func openAtIndex(lg *zap.Logger, dirpath string, snap walpb.Snapshot, write bool) (*WAL, error) { |
| 333 | names, nameIndex, err := selectWALFiles(lg, dirpath, snap) |
| 334 | if err != nil { |
| 335 | return nil, err |
| 336 | } |
| 337 | |
| 338 | rs, ls, closer, err := openWALFiles(lg, dirpath, names, nameIndex, write) |
| 339 | if err != nil { |
| 340 | return nil, err |
| 341 | } |
| 342 | |
| 343 | // create a WAL ready for reading |
| 344 | w := &WAL{ |
| 345 | lg: lg, |
| 346 | dir: dirpath, |
| 347 | start: snap, |
| 348 | decoder: newDecoder(rs...), |
| 349 | readClose: closer, |
| 350 | locks: ls, |
| 351 | } |
| 352 | |
| 353 | if write { |
| 354 | // write reuses the file descriptors from read; don't close so |
| 355 | // WAL can append without dropping the file lock |
| 356 | w.readClose = nil |
| 357 | if _, _, err := parseWALName(filepath.Base(w.tail().Name())); err != nil { |
| 358 | closer() |
| 359 | return nil, err |
| 360 | } |
| 361 | w.fp = newFilePipeline(lg, w.dir, SegmentSizeBytes) |
| 362 | } |
| 363 | |
| 364 | return w, nil |
| 365 | } |
| 366 | |
| 367 | func selectWALFiles(lg *zap.Logger, dirpath string, snap walpb.Snapshot) ([]string, int, error) { |
| 368 | names, err := readWALNames(lg, dirpath) |
| 369 | if err != nil { |
| 370 | return nil, -1, err |
| 371 | } |
| 372 | |
| 373 | nameIndex, ok := searchIndex(lg, names, snap.Index) |
| 374 | if !ok || !isValidSeq(lg, names[nameIndex:]) { |
| 375 | err = ErrFileNotFound |
| 376 | return nil, -1, err |
| 377 | } |
| 378 | |
| 379 | return names, nameIndex, nil |
| 380 | } |
| 381 | |
| 382 | func openWALFiles(lg *zap.Logger, dirpath string, names []string, nameIndex int, write bool) ([]io.Reader, []*fileutil.LockedFile, func() error, error) { |
| 383 | rcs := make([]io.ReadCloser, 0) |
| 384 | rs := make([]io.Reader, 0) |
| 385 | ls := make([]*fileutil.LockedFile, 0) |
| 386 | for _, name := range names[nameIndex:] { |
| 387 | p := filepath.Join(dirpath, name) |
| 388 | if write { |
| 389 | l, err := fileutil.TryLockFile(p, os.O_RDWR, fileutil.PrivateFileMode) |
| 390 | if err != nil { |
| 391 | closeAll(rcs...) |
| 392 | return nil, nil, nil, err |
| 393 | } |
| 394 | ls = append(ls, l) |
| 395 | rcs = append(rcs, l) |
| 396 | } else { |
| 397 | rf, err := os.OpenFile(p, os.O_RDONLY, fileutil.PrivateFileMode) |
| 398 | if err != nil { |
| 399 | closeAll(rcs...) |
| 400 | return nil, nil, nil, err |
| 401 | } |
| 402 | ls = append(ls, nil) |
| 403 | rcs = append(rcs, rf) |
| 404 | } |
| 405 | rs = append(rs, rcs[len(rcs)-1]) |
| 406 | } |
| 407 | |
| 408 | closer := func() error { return closeAll(rcs...) } |
| 409 | |
| 410 | return rs, ls, closer, nil |
| 411 | } |
| 412 | |
| 413 | // ReadAll reads out records of the current WAL. |
| 414 | // If opened in write mode, it must read out all records until EOF. Or an error |
| 415 | // will be returned. |
| 416 | // If opened in read mode, it will try to read all records if possible. |
| 417 | // If it cannot read out the expected snap, it will return ErrSnapshotNotFound. |
| 418 | // If loaded snap doesn't match with the expected one, it will return |
| 419 | // all the records and error ErrSnapshotMismatch. |
| 420 | // TODO: detect not-last-snap error. |
| 421 | // TODO: maybe loose the checking of match. |
| 422 | // After ReadAll, the WAL will be ready for appending new records. |
| 423 | func (w *WAL) ReadAll() (metadata []byte, state raftpb.HardState, ents []raftpb.Entry, err error) { |
| 424 | w.mu.Lock() |
| 425 | defer w.mu.Unlock() |
| 426 | |
| 427 | rec := &walpb.Record{} |
| 428 | decoder := w.decoder |
| 429 | |
| 430 | var match bool |
| 431 | for err = decoder.decode(rec); err == nil; err = decoder.decode(rec) { |
| 432 | switch rec.Type { |
| 433 | case entryType: |
| 434 | e := mustUnmarshalEntry(rec.Data) |
| 435 | if e.Index > w.start.Index { |
| 436 | ents = append(ents[:e.Index-w.start.Index-1], e) |
| 437 | } |
| 438 | w.enti = e.Index |
| 439 | |
| 440 | case stateType: |
| 441 | state = mustUnmarshalState(rec.Data) |
| 442 | |
| 443 | case metadataType: |
| 444 | if metadata != nil && !bytes.Equal(metadata, rec.Data) { |
| 445 | state.Reset() |
| 446 | return nil, state, nil, ErrMetadataConflict |
| 447 | } |
| 448 | metadata = rec.Data |
| 449 | |
| 450 | case crcType: |
| 451 | crc := decoder.crc.Sum32() |
| 452 | // current crc of decoder must match the crc of the record. |
| 453 | // do no need to match 0 crc, since the decoder is a new one at this case. |
| 454 | if crc != 0 && rec.Validate(crc) != nil { |
| 455 | state.Reset() |
| 456 | return nil, state, nil, ErrCRCMismatch |
| 457 | } |
| 458 | decoder.updateCRC(rec.Crc) |
| 459 | |
| 460 | case snapshotType: |
| 461 | var snap walpb.Snapshot |
| 462 | pbutil.MustUnmarshal(&snap, rec.Data) |
| 463 | if snap.Index == w.start.Index { |
| 464 | if snap.Term != w.start.Term { |
| 465 | state.Reset() |
| 466 | return nil, state, nil, ErrSnapshotMismatch |
| 467 | } |
| 468 | match = true |
| 469 | } |
| 470 | |
| 471 | default: |
| 472 | state.Reset() |
| 473 | return nil, state, nil, fmt.Errorf("unexpected block type %d", rec.Type) |
| 474 | } |
| 475 | } |
| 476 | |
| 477 | switch w.tail() { |
| 478 | case nil: |
| 479 | // We do not have to read out all entries in read mode. |
| 480 | // The last record maybe a partial written one, so |
| 481 | // ErrunexpectedEOF might be returned. |
| 482 | if err != io.EOF && err != io.ErrUnexpectedEOF { |
| 483 | state.Reset() |
| 484 | return nil, state, nil, err |
| 485 | } |
| 486 | default: |
| 487 | // We must read all of the entries if WAL is opened in write mode. |
| 488 | if err != io.EOF { |
| 489 | state.Reset() |
| 490 | return nil, state, nil, err |
| 491 | } |
| 492 | // decodeRecord() will return io.EOF if it detects a zero record, |
| 493 | // but this zero record may be followed by non-zero records from |
| 494 | // a torn write. Overwriting some of these non-zero records, but |
| 495 | // not all, will cause CRC errors on WAL open. Since the records |
| 496 | // were never fully synced to disk in the first place, it's safe |
| 497 | // to zero them out to avoid any CRC errors from new writes. |
| 498 | if _, err = w.tail().Seek(w.decoder.lastOffset(), io.SeekStart); err != nil { |
| 499 | return nil, state, nil, err |
| 500 | } |
| 501 | if err = fileutil.ZeroToEnd(w.tail().File); err != nil { |
| 502 | return nil, state, nil, err |
| 503 | } |
| 504 | } |
| 505 | |
| 506 | err = nil |
| 507 | if !match { |
| 508 | err = ErrSnapshotNotFound |
| 509 | } |
| 510 | |
| 511 | // close decoder, disable reading |
| 512 | if w.readClose != nil { |
| 513 | w.readClose() |
| 514 | w.readClose = nil |
| 515 | } |
| 516 | w.start = walpb.Snapshot{} |
| 517 | |
| 518 | w.metadata = metadata |
| 519 | |
| 520 | if w.tail() != nil { |
| 521 | // create encoder (chain crc with the decoder), enable appending |
| 522 | w.encoder, err = newFileEncoder(w.tail().File, w.decoder.lastCRC()) |
| 523 | if err != nil { |
| 524 | return |
| 525 | } |
| 526 | } |
| 527 | w.decoder = nil |
| 528 | |
| 529 | return metadata, state, ents, err |
| 530 | } |
| 531 | |
| 532 | // Verify reads through the given WAL and verifies that it is not corrupted. |
| 533 | // It creates a new decoder to read through the records of the given WAL. |
| 534 | // It does not conflict with any open WAL, but it is recommended not to |
| 535 | // call this function after opening the WAL for writing. |
| 536 | // If it cannot read out the expected snap, it will return ErrSnapshotNotFound. |
| 537 | // If the loaded snap doesn't match with the expected one, it will |
| 538 | // return error ErrSnapshotMismatch. |
| 539 | func Verify(lg *zap.Logger, walDir string, snap walpb.Snapshot) error { |
| 540 | var metadata []byte |
| 541 | var err error |
| 542 | var match bool |
| 543 | |
| 544 | rec := &walpb.Record{} |
| 545 | |
| 546 | names, nameIndex, err := selectWALFiles(lg, walDir, snap) |
| 547 | if err != nil { |
| 548 | return err |
| 549 | } |
| 550 | |
| 551 | // open wal files in read mode, so that there is no conflict |
| 552 | // when the same WAL is opened elsewhere in write mode |
| 553 | rs, _, closer, err := openWALFiles(lg, walDir, names, nameIndex, false) |
| 554 | if err != nil { |
| 555 | return err |
| 556 | } |
| 557 | defer func() { |
| 558 | if closer != nil { |
| 559 | closer() |
| 560 | } |
| 561 | }() |
| 562 | |
| 563 | // create a new decoder from the readers on the WAL files |
| 564 | decoder := newDecoder(rs...) |
| 565 | |
| 566 | for err = decoder.decode(rec); err == nil; err = decoder.decode(rec) { |
| 567 | switch rec.Type { |
| 568 | case metadataType: |
| 569 | if metadata != nil && !bytes.Equal(metadata, rec.Data) { |
| 570 | return ErrMetadataConflict |
| 571 | } |
| 572 | metadata = rec.Data |
| 573 | case crcType: |
| 574 | crc := decoder.crc.Sum32() |
| 575 | // Current crc of decoder must match the crc of the record. |
| 576 | // We need not match 0 crc, since the decoder is a new one at this point. |
| 577 | if crc != 0 && rec.Validate(crc) != nil { |
| 578 | return ErrCRCMismatch |
| 579 | } |
| 580 | decoder.updateCRC(rec.Crc) |
| 581 | case snapshotType: |
| 582 | var loadedSnap walpb.Snapshot |
| 583 | pbutil.MustUnmarshal(&loadedSnap, rec.Data) |
| 584 | if loadedSnap.Index == snap.Index { |
| 585 | if loadedSnap.Term != snap.Term { |
| 586 | return ErrSnapshotMismatch |
| 587 | } |
| 588 | match = true |
| 589 | } |
| 590 | // We ignore all entry and state type records as these |
| 591 | // are not necessary for validating the WAL contents |
| 592 | case entryType: |
| 593 | case stateType: |
| 594 | default: |
| 595 | return fmt.Errorf("unexpected block type %d", rec.Type) |
| 596 | } |
| 597 | } |
| 598 | |
| 599 | // We do not have to read out all the WAL entries |
| 600 | // as the decoder is opened in read mode. |
| 601 | if err != io.EOF && err != io.ErrUnexpectedEOF { |
| 602 | return err |
| 603 | } |
| 604 | |
| 605 | if !match { |
| 606 | return ErrSnapshotNotFound |
| 607 | } |
| 608 | |
| 609 | return nil |
| 610 | } |
| 611 | |
| 612 | // cut closes current file written and creates a new one ready to append. |
| 613 | // cut first creates a temp wal file and writes necessary headers into it. |
| 614 | // Then cut atomically rename temp wal file to a wal file. |
| 615 | func (w *WAL) cut() error { |
| 616 | // close old wal file; truncate to avoid wasting space if an early cut |
| 617 | off, serr := w.tail().Seek(0, io.SeekCurrent) |
| 618 | if serr != nil { |
| 619 | return serr |
| 620 | } |
| 621 | |
| 622 | if err := w.tail().Truncate(off); err != nil { |
| 623 | return err |
| 624 | } |
| 625 | |
| 626 | if err := w.sync(); err != nil { |
| 627 | return err |
| 628 | } |
| 629 | |
| 630 | fpath := filepath.Join(w.dir, walName(w.seq()+1, w.enti+1)) |
| 631 | |
| 632 | // create a temp wal file with name sequence + 1, or truncate the existing one |
| 633 | newTail, err := w.fp.Open() |
| 634 | if err != nil { |
| 635 | return err |
| 636 | } |
| 637 | |
| 638 | // update writer and save the previous crc |
| 639 | w.locks = append(w.locks, newTail) |
| 640 | prevCrc := w.encoder.crc.Sum32() |
| 641 | w.encoder, err = newFileEncoder(w.tail().File, prevCrc) |
| 642 | if err != nil { |
| 643 | return err |
| 644 | } |
| 645 | |
| 646 | if err = w.saveCrc(prevCrc); err != nil { |
| 647 | return err |
| 648 | } |
| 649 | |
| 650 | if err = w.encoder.encode(&walpb.Record{Type: metadataType, Data: w.metadata}); err != nil { |
| 651 | return err |
| 652 | } |
| 653 | |
| 654 | if err = w.saveState(&w.state); err != nil { |
| 655 | return err |
| 656 | } |
| 657 | |
| 658 | // atomically move temp wal file to wal file |
| 659 | if err = w.sync(); err != nil { |
| 660 | return err |
| 661 | } |
| 662 | |
| 663 | off, err = w.tail().Seek(0, io.SeekCurrent) |
| 664 | if err != nil { |
| 665 | return err |
| 666 | } |
| 667 | |
| 668 | if err = os.Rename(newTail.Name(), fpath); err != nil { |
| 669 | return err |
| 670 | } |
| 671 | if err = fileutil.Fsync(w.dirFile); err != nil { |
| 672 | return err |
| 673 | } |
| 674 | |
| 675 | // reopen newTail with its new path so calls to Name() match the wal filename format |
| 676 | newTail.Close() |
| 677 | |
| 678 | if newTail, err = fileutil.LockFile(fpath, os.O_WRONLY, fileutil.PrivateFileMode); err != nil { |
| 679 | return err |
| 680 | } |
| 681 | if _, err = newTail.Seek(off, io.SeekStart); err != nil { |
| 682 | return err |
| 683 | } |
| 684 | |
| 685 | w.locks[len(w.locks)-1] = newTail |
| 686 | |
| 687 | prevCrc = w.encoder.crc.Sum32() |
| 688 | w.encoder, err = newFileEncoder(w.tail().File, prevCrc) |
| 689 | if err != nil { |
| 690 | return err |
| 691 | } |
| 692 | |
| 693 | if w.lg != nil { |
| 694 | w.lg.Info("created a new WAL segment", zap.String("path", fpath)) |
| 695 | } else { |
| 696 | plog.Infof("segmented wal file %v is created", fpath) |
| 697 | } |
| 698 | return nil |
| 699 | } |
| 700 | |
| 701 | func (w *WAL) sync() error { |
| 702 | if w.encoder != nil { |
| 703 | if err := w.encoder.flush(); err != nil { |
| 704 | return err |
| 705 | } |
| 706 | } |
| 707 | start := time.Now() |
| 708 | err := fileutil.Fdatasync(w.tail().File) |
| 709 | |
| 710 | took := time.Since(start) |
| 711 | if took > warnSyncDuration { |
| 712 | if w.lg != nil { |
| 713 | w.lg.Warn( |
| 714 | "slow fdatasync", |
| 715 | zap.Duration("took", took), |
| 716 | zap.Duration("expected-duration", warnSyncDuration), |
| 717 | ) |
| 718 | } else { |
| 719 | plog.Warningf("sync duration of %v, expected less than %v", took, warnSyncDuration) |
| 720 | } |
| 721 | } |
| 722 | walFsyncSec.Observe(took.Seconds()) |
| 723 | |
| 724 | return err |
| 725 | } |
| 726 | |
| 727 | // ReleaseLockTo releases the locks, which has smaller index than the given index |
| 728 | // except the largest one among them. |
| 729 | // For example, if WAL is holding lock 1,2,3,4,5,6, ReleaseLockTo(4) will release |
| 730 | // lock 1,2 but keep 3. ReleaseLockTo(5) will release 1,2,3 but keep 4. |
| 731 | func (w *WAL) ReleaseLockTo(index uint64) error { |
| 732 | w.mu.Lock() |
| 733 | defer w.mu.Unlock() |
| 734 | |
| 735 | if len(w.locks) == 0 { |
| 736 | return nil |
| 737 | } |
| 738 | |
| 739 | var smaller int |
| 740 | found := false |
| 741 | for i, l := range w.locks { |
| 742 | _, lockIndex, err := parseWALName(filepath.Base(l.Name())) |
| 743 | if err != nil { |
| 744 | return err |
| 745 | } |
| 746 | if lockIndex >= index { |
| 747 | smaller = i - 1 |
| 748 | found = true |
| 749 | break |
| 750 | } |
| 751 | } |
| 752 | |
| 753 | // if no lock index is greater than the release index, we can |
| 754 | // release lock up to the last one(excluding). |
| 755 | if !found { |
| 756 | smaller = len(w.locks) - 1 |
| 757 | } |
| 758 | |
| 759 | if smaller <= 0 { |
| 760 | return nil |
| 761 | } |
| 762 | |
| 763 | for i := 0; i < smaller; i++ { |
| 764 | if w.locks[i] == nil { |
| 765 | continue |
| 766 | } |
| 767 | w.locks[i].Close() |
| 768 | } |
| 769 | w.locks = w.locks[smaller:] |
| 770 | |
| 771 | return nil |
| 772 | } |
| 773 | |
| 774 | // Close closes the current WAL file and directory. |
| 775 | func (w *WAL) Close() error { |
| 776 | w.mu.Lock() |
| 777 | defer w.mu.Unlock() |
| 778 | |
| 779 | if w.fp != nil { |
| 780 | w.fp.Close() |
| 781 | w.fp = nil |
| 782 | } |
| 783 | |
| 784 | if w.tail() != nil { |
| 785 | if err := w.sync(); err != nil { |
| 786 | return err |
| 787 | } |
| 788 | } |
| 789 | for _, l := range w.locks { |
| 790 | if l == nil { |
| 791 | continue |
| 792 | } |
| 793 | if err := l.Close(); err != nil { |
| 794 | if w.lg != nil { |
| 795 | w.lg.Warn("failed to close WAL", zap.Error(err)) |
| 796 | } else { |
| 797 | plog.Errorf("failed to unlock during closing wal: %s", err) |
| 798 | } |
| 799 | } |
| 800 | } |
| 801 | |
| 802 | return w.dirFile.Close() |
| 803 | } |
| 804 | |
| 805 | func (w *WAL) saveEntry(e *raftpb.Entry) error { |
| 806 | // TODO: add MustMarshalTo to reduce one allocation. |
| 807 | b := pbutil.MustMarshal(e) |
| 808 | rec := &walpb.Record{Type: entryType, Data: b} |
| 809 | if err := w.encoder.encode(rec); err != nil { |
| 810 | return err |
| 811 | } |
| 812 | w.enti = e.Index |
| 813 | return nil |
| 814 | } |
| 815 | |
| 816 | func (w *WAL) saveState(s *raftpb.HardState) error { |
| 817 | if raft.IsEmptyHardState(*s) { |
| 818 | return nil |
| 819 | } |
| 820 | w.state = *s |
| 821 | b := pbutil.MustMarshal(s) |
| 822 | rec := &walpb.Record{Type: stateType, Data: b} |
| 823 | return w.encoder.encode(rec) |
| 824 | } |
| 825 | |
| 826 | func (w *WAL) Save(st raftpb.HardState, ents []raftpb.Entry) error { |
| 827 | w.mu.Lock() |
| 828 | defer w.mu.Unlock() |
| 829 | |
| 830 | // short cut, do not call sync |
| 831 | if raft.IsEmptyHardState(st) && len(ents) == 0 { |
| 832 | return nil |
| 833 | } |
| 834 | |
| 835 | mustSync := raft.MustSync(st, w.state, len(ents)) |
| 836 | |
| 837 | // TODO(xiangli): no more reference operator |
| 838 | for i := range ents { |
| 839 | if err := w.saveEntry(&ents[i]); err != nil { |
| 840 | return err |
| 841 | } |
| 842 | } |
| 843 | if err := w.saveState(&st); err != nil { |
| 844 | return err |
| 845 | } |
| 846 | |
| 847 | curOff, err := w.tail().Seek(0, io.SeekCurrent) |
| 848 | if err != nil { |
| 849 | return err |
| 850 | } |
| 851 | if curOff < SegmentSizeBytes { |
| 852 | if mustSync { |
| 853 | return w.sync() |
| 854 | } |
| 855 | return nil |
| 856 | } |
| 857 | |
| 858 | return w.cut() |
| 859 | } |
| 860 | |
| 861 | func (w *WAL) SaveSnapshot(e walpb.Snapshot) error { |
| 862 | b := pbutil.MustMarshal(&e) |
| 863 | |
| 864 | w.mu.Lock() |
| 865 | defer w.mu.Unlock() |
| 866 | |
| 867 | rec := &walpb.Record{Type: snapshotType, Data: b} |
| 868 | if err := w.encoder.encode(rec); err != nil { |
| 869 | return err |
| 870 | } |
| 871 | // update enti only when snapshot is ahead of last index |
| 872 | if w.enti < e.Index { |
| 873 | w.enti = e.Index |
| 874 | } |
| 875 | return w.sync() |
| 876 | } |
| 877 | |
| 878 | func (w *WAL) saveCrc(prevCrc uint32) error { |
| 879 | return w.encoder.encode(&walpb.Record{Type: crcType, Crc: prevCrc}) |
| 880 | } |
| 881 | |
| 882 | func (w *WAL) tail() *fileutil.LockedFile { |
| 883 | if len(w.locks) > 0 { |
| 884 | return w.locks[len(w.locks)-1] |
| 885 | } |
| 886 | return nil |
| 887 | } |
| 888 | |
| 889 | func (w *WAL) seq() uint64 { |
| 890 | t := w.tail() |
| 891 | if t == nil { |
| 892 | return 0 |
| 893 | } |
| 894 | seq, _, err := parseWALName(filepath.Base(t.Name())) |
| 895 | if err != nil { |
| 896 | if w.lg != nil { |
| 897 | w.lg.Fatal("failed to parse WAL name", zap.String("name", t.Name()), zap.Error(err)) |
| 898 | } else { |
| 899 | plog.Fatalf("bad wal name %s (%v)", t.Name(), err) |
| 900 | } |
| 901 | } |
| 902 | return seq |
| 903 | } |
| 904 | |
| 905 | func closeAll(rcs ...io.ReadCloser) error { |
| 906 | for _, f := range rcs { |
| 907 | if err := f.Close(); err != nil { |
| 908 | return err |
| 909 | } |
| 910 | } |
| 911 | return nil |
| 912 | } |