wal.go 7.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305
  1. /*
  2. Copyright 2014 CoreOS, Inc.
  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. http://www.apache.org/licenses/LICENSE-2.0
  7. Unless required by applicable law or agreed to in writing, software
  8. distributed under the License is distributed on an "AS IS" BASIS,
  9. WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  10. See the License for the specific language governing permissions and
  11. limitations under the License.
  12. */
  13. package wal
  14. import (
  15. "errors"
  16. "fmt"
  17. "hash/crc32"
  18. "io"
  19. "os"
  20. "path"
  21. "reflect"
  22. "sort"
  23. "github.com/coreos/etcd/pkg/fileutil"
  24. "github.com/coreos/etcd/pkg/pbutil"
  25. "github.com/coreos/etcd/raft"
  26. "github.com/coreos/etcd/raft/raftpb"
  27. "github.com/coreos/etcd/wal/walpb"
  28. )
  29. const (
  30. metadataType int64 = iota + 1
  31. entryType
  32. stateType
  33. crcType
  34. // the owner can make/remove files inside the directory
  35. privateDirMode = 0700
  36. )
  37. var (
  38. ErrMetadataConflict = errors.New("wal: conflicting metadata found")
  39. ErrFileNotFound = errors.New("wal: file not found")
  40. ErrCRCMismatch = errors.New("wal: crc mismatch")
  41. crcTable = crc32.MakeTable(crc32.Castagnoli)
  42. )
  43. // WAL is a logical repersentation of the stable storage.
  44. // WAL is either in read mode or append mode but not both.
  45. // A newly created WAL is in append mode, and ready for appending records.
  46. // A just opened WAL is in read mode, and ready for reading records.
  47. // The WAL will be ready for appending after reading out all the previous records.
  48. type WAL struct {
  49. dir string // the living directory of the underlay files
  50. metadata []byte // metadata recorded at the head of each WAL
  51. state raftpb.HardState // hardstate recorded at the head of WAL
  52. ri uint64 // index of entry to start reading
  53. decoder *decoder // decoder to decode records
  54. f *os.File // underlay file opened for appending, sync
  55. seq uint64 // sequence of the wal file currently used for writes
  56. enti uint64 // index of the last entry saved to the wal
  57. encoder *encoder // encoder to encode records
  58. }
  59. // Create creates a WAL ready for appending records. The given metadata is
  60. // recorded at the head of each WAL file, and can be retrieved with ReadAll.
  61. func Create(dirpath string, metadata []byte) (*WAL, error) {
  62. if Exist(dirpath) {
  63. return nil, os.ErrExist
  64. }
  65. if err := os.MkdirAll(dirpath, privateDirMode); err != nil {
  66. return nil, err
  67. }
  68. p := path.Join(dirpath, walName(0, 0))
  69. f, err := os.OpenFile(p, os.O_WRONLY|os.O_APPEND|os.O_CREATE, 0600)
  70. if err != nil {
  71. return nil, err
  72. }
  73. w := &WAL{
  74. dir: dirpath,
  75. metadata: metadata,
  76. seq: 0,
  77. f: f,
  78. encoder: newEncoder(f, 0),
  79. }
  80. if err := w.saveCrc(0); err != nil {
  81. return nil, err
  82. }
  83. if err := w.encoder.encode(&walpb.Record{Type: metadataType, Data: metadata}); err != nil {
  84. return nil, err
  85. }
  86. if err = w.sync(); err != nil {
  87. return nil, err
  88. }
  89. return w, nil
  90. }
  91. // OpenAtIndex opens the WAL at the given index.
  92. // The index SHOULD have been previously committed to the WAL, or the following
  93. // ReadAll will fail.
  94. // The returned WAL is ready to read and the first record will be the given
  95. // index. The WAL cannot be appended to before reading out all of its
  96. // previous records.
  97. func OpenAtIndex(dirpath string, index uint64) (*WAL, error) {
  98. names, err := fileutil.ReadDir(dirpath)
  99. if err != nil {
  100. return nil, err
  101. }
  102. names = checkWalNames(names)
  103. if len(names) == 0 {
  104. return nil, ErrFileNotFound
  105. }
  106. sort.Sort(sort.StringSlice(names))
  107. nameIndex, ok := searchIndex(names, index)
  108. if !ok || !isValidSeq(names[nameIndex:]) {
  109. return nil, ErrFileNotFound
  110. }
  111. // open the wal files for reading
  112. rcs := make([]io.ReadCloser, 0)
  113. for _, name := range names[nameIndex:] {
  114. f, err := os.Open(path.Join(dirpath, name))
  115. if err != nil {
  116. return nil, err
  117. }
  118. rcs = append(rcs, f)
  119. }
  120. rc := MultiReadCloser(rcs...)
  121. // open the lastest wal file for appending
  122. seq, _, err := parseWalName(names[len(names)-1])
  123. if err != nil {
  124. rc.Close()
  125. return nil, err
  126. }
  127. last := path.Join(dirpath, names[len(names)-1])
  128. f, err := os.OpenFile(last, os.O_WRONLY|os.O_APPEND, 0)
  129. if err != nil {
  130. rc.Close()
  131. return nil, err
  132. }
  133. // create a WAL ready for reading
  134. w := &WAL{
  135. dir: dirpath,
  136. ri: index,
  137. decoder: newDecoder(rc),
  138. f: f,
  139. seq: seq,
  140. }
  141. return w, nil
  142. }
  143. // ReadAll reads out all records of the current WAL.
  144. // If it cannot read out the expected entry, it will return ErrIndexNotFound.
  145. // After ReadAll, the WAL will be ready for appending new records.
  146. func (w *WAL) ReadAll() (metadata []byte, state raftpb.HardState, ents []raftpb.Entry, err error) {
  147. rec := &walpb.Record{}
  148. decoder := w.decoder
  149. for err = decoder.decode(rec); err == nil; err = decoder.decode(rec) {
  150. switch rec.Type {
  151. case entryType:
  152. e := mustUnmarshalEntry(rec.Data)
  153. if e.Index >= w.ri {
  154. ents = append(ents[:e.Index-w.ri], e)
  155. }
  156. w.enti = e.Index
  157. case stateType:
  158. state = mustUnmarshalState(rec.Data)
  159. case metadataType:
  160. if metadata != nil && !reflect.DeepEqual(metadata, rec.Data) {
  161. state.Reset()
  162. return nil, state, nil, ErrMetadataConflict
  163. }
  164. metadata = rec.Data
  165. case crcType:
  166. crc := decoder.crc.Sum32()
  167. // current crc of decoder must match the crc of the record.
  168. // do no need to match 0 crc, since the decoder is a new one at this case.
  169. if crc != 0 && rec.Validate(crc) != nil {
  170. state.Reset()
  171. return nil, state, nil, ErrCRCMismatch
  172. }
  173. decoder.updateCRC(rec.Crc)
  174. default:
  175. state.Reset()
  176. return nil, state, nil, fmt.Errorf("unexpected block type %d", rec.Type)
  177. }
  178. }
  179. if err != io.EOF {
  180. state.Reset()
  181. return nil, state, nil, err
  182. }
  183. // close decoder, disable reading
  184. w.decoder.close()
  185. w.ri = 0
  186. w.metadata = metadata
  187. // create encoder (chain crc with the decoder), enable appending
  188. w.encoder = newEncoder(w.f, w.decoder.lastCRC())
  189. w.decoder = nil
  190. return metadata, state, ents, nil
  191. }
  192. // Cut closes current file written and creates a new one ready to append.
  193. func (w *WAL) Cut() error {
  194. // create a new wal file with name sequence + 1
  195. fpath := path.Join(w.dir, walName(w.seq+1, w.enti+1))
  196. f, err := os.OpenFile(fpath, os.O_WRONLY|os.O_APPEND|os.O_CREATE, 0600)
  197. if err != nil {
  198. return err
  199. }
  200. if err = w.sync(); err != nil {
  201. return err
  202. }
  203. w.f.Close()
  204. // update writer and save the previous crc
  205. w.f = f
  206. w.seq++
  207. prevCrc := w.encoder.crc.Sum32()
  208. w.encoder = newEncoder(w.f, prevCrc)
  209. if err := w.saveCrc(prevCrc); err != nil {
  210. return err
  211. }
  212. if err := w.encoder.encode(&walpb.Record{Type: metadataType, Data: w.metadata}); err != nil {
  213. return err
  214. }
  215. if err := w.SaveState(&w.state); err != nil {
  216. return err
  217. }
  218. return w.sync()
  219. }
  220. func (w *WAL) sync() error {
  221. if w.encoder != nil {
  222. if err := w.encoder.flush(); err != nil {
  223. return err
  224. }
  225. }
  226. return w.f.Sync()
  227. }
  228. func (w *WAL) Close() error {
  229. if w.f != nil {
  230. if err := w.sync(); err != nil {
  231. return err
  232. }
  233. if err := w.f.Close(); err != nil {
  234. return err
  235. }
  236. }
  237. return nil
  238. }
  239. func (w *WAL) SaveEntry(e *raftpb.Entry) error {
  240. b := pbutil.MustMarshal(e)
  241. rec := &walpb.Record{Type: entryType, Data: b}
  242. if err := w.encoder.encode(rec); err != nil {
  243. return err
  244. }
  245. w.enti = e.Index
  246. return nil
  247. }
  248. func (w *WAL) SaveState(s *raftpb.HardState) error {
  249. if raft.IsEmptyHardState(*s) {
  250. return nil
  251. }
  252. w.state = *s
  253. b := pbutil.MustMarshal(s)
  254. rec := &walpb.Record{Type: stateType, Data: b}
  255. return w.encoder.encode(rec)
  256. }
  257. func (w *WAL) Save(st raftpb.HardState, ents []raftpb.Entry) error {
  258. // TODO(xiangli): no more reference operator
  259. if err := w.SaveState(&st); err != nil {
  260. return err
  261. }
  262. for i := range ents {
  263. if err := w.SaveEntry(&ents[i]); err != nil {
  264. return err
  265. }
  266. }
  267. return w.sync()
  268. }
  269. func (w *WAL) saveCrc(prevCrc uint32) error {
  270. return w.encoder.encode(&walpb.Record{Type: crcType, Crc: prevCrc})
  271. }