wal.go 7.1 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291
  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. "log"
  20. "os"
  21. "path"
  22. "sort"
  23. "github.com/coreos/etcd/raft"
  24. "github.com/coreos/etcd/raft/raftpb"
  25. "github.com/coreos/etcd/wal/walpb"
  26. )
  27. const (
  28. infoType int64 = iota + 1
  29. entryType
  30. stateType
  31. crcType
  32. // the owner can make/remove files inside the directory
  33. privateDirMode = 0700
  34. )
  35. var (
  36. ErrIDMismatch = errors.New("wal: unmatch id")
  37. ErrNotFound = errors.New("wal: file is not found")
  38. ErrCRCMismatch = errors.New("wal: crc mismatch")
  39. crcTable = crc32.MakeTable(crc32.Castagnoli)
  40. )
  41. // WAL is a logical repersentation of the stable storage.
  42. // WAL is either in read mode or append mode but not both.
  43. // A newly created WAL is in append mode, and ready for appending records.
  44. // A just opened WAL is in read mode, and ready for reading records.
  45. // The WAL will be ready for appending after reading out all the previous records.
  46. type WAL struct {
  47. dir string // the living directory of the underlay files
  48. ri int64 // index of entry to start reading
  49. decoder *decoder // decoder to decode records
  50. f *os.File // underlay file opened for appending, sync
  51. seq int64 // sequence of the wal file currently used for writes
  52. enti int64 // index of the last entry saved to the wal
  53. encoder *encoder // encoder to encode records
  54. }
  55. // Create creates a WAL ready for appending records.
  56. func Create(dirpath string) (*WAL, error) {
  57. log.Printf("path=%s wal.create", dirpath)
  58. if Exist(dirpath) {
  59. return nil, os.ErrExist
  60. }
  61. if err := os.MkdirAll(dirpath, privateDirMode); err != nil {
  62. return nil, err
  63. }
  64. p := path.Join(dirpath, walName(0, 0))
  65. f, err := os.OpenFile(p, os.O_WRONLY|os.O_APPEND|os.O_CREATE, 0600)
  66. if err != nil {
  67. return nil, err
  68. }
  69. w := &WAL{
  70. dir: dirpath,
  71. seq: 0,
  72. f: f,
  73. encoder: newEncoder(f, 0),
  74. }
  75. if err := w.saveCrc(0); err != nil {
  76. return nil, err
  77. }
  78. return w, nil
  79. }
  80. // OpenAtIndex opens the WAL at the given index.
  81. // The index MUST have been previously committed to the WAL.
  82. // The returned WAL is ready to read and the first record will be the given
  83. // index. The WAL cannot be appended to before reading out all of its
  84. // previous records.
  85. func OpenAtIndex(dirpath string, index int64) (*WAL, error) {
  86. log.Printf("path=%s wal.load index=%d", dirpath, index)
  87. names, err := readDir(dirpath)
  88. if err != nil {
  89. return nil, err
  90. }
  91. names = checkWalNames(names)
  92. if len(names) == 0 {
  93. return nil, ErrNotFound
  94. }
  95. sort.Sort(sort.StringSlice(names))
  96. nameIndex, ok := searchIndex(names, index)
  97. if !ok || !isValidSeq(names[nameIndex:]) {
  98. return nil, ErrNotFound
  99. }
  100. // open the wal files for reading
  101. rcs := make([]io.ReadCloser, 0)
  102. for _, name := range names[nameIndex:] {
  103. f, err := os.Open(path.Join(dirpath, name))
  104. if err != nil {
  105. return nil, err
  106. }
  107. rcs = append(rcs, f)
  108. }
  109. rc := MultiReadCloser(rcs...)
  110. // open the lastest wal file for appending
  111. seq, _, err := parseWalName(names[len(names)-1])
  112. if err != nil {
  113. rc.Close()
  114. return nil, err
  115. }
  116. last := path.Join(dirpath, names[len(names)-1])
  117. f, err := os.OpenFile(last, os.O_WRONLY|os.O_APPEND, 0)
  118. if err != nil {
  119. rc.Close()
  120. return nil, err
  121. }
  122. // create a WAL ready for reading
  123. w := &WAL{
  124. ri: index,
  125. decoder: newDecoder(rc),
  126. f: f,
  127. seq: seq,
  128. }
  129. return w, nil
  130. }
  131. // ReadAll reads out all records of the current WAL.
  132. // After ReadAll, the WAL will be ready for appending new records.
  133. func (w *WAL) ReadAll() (id int64, state raftpb.HardState, ents []raftpb.Entry, err error) {
  134. rec := &walpb.Record{}
  135. decoder := w.decoder
  136. for err = decoder.decode(rec); err == nil; err = decoder.decode(rec) {
  137. switch rec.Type {
  138. case entryType:
  139. e := mustUnmarshalEntry(rec.Data)
  140. if e.Index >= w.ri {
  141. ents = append(ents[:e.Index-w.ri], e)
  142. }
  143. w.enti = e.Index
  144. case stateType:
  145. state = mustUnmarshalState(rec.Data)
  146. case infoType:
  147. i := mustUnmarshalInfo(rec.Data)
  148. if id != 0 && id != i.Id {
  149. state.Reset()
  150. return 0, state, nil, ErrIDMismatch
  151. }
  152. id = i.Id
  153. case crcType:
  154. crc := decoder.crc.Sum32()
  155. // current crc of decoder must match the crc of the record.
  156. // do no need to match 0 crc, since the decoder is a new one at this case.
  157. if crc != 0 && rec.Validate(crc) != nil {
  158. state.Reset()
  159. return 0, state, nil, ErrCRCMismatch
  160. }
  161. decoder.updateCRC(rec.Crc)
  162. default:
  163. state.Reset()
  164. return 0, state, nil, fmt.Errorf("unexpected block type %d", rec.Type)
  165. }
  166. }
  167. if err != io.EOF {
  168. state.Reset()
  169. return 0, state, nil, err
  170. }
  171. // close decoder, disable reading
  172. w.decoder.close()
  173. w.ri = 0
  174. // create encoder (chain crc with the decoder), enable appending
  175. w.encoder = newEncoder(w.f, w.decoder.lastCRC())
  176. w.decoder = nil
  177. return id, state, ents, nil
  178. }
  179. // Cut closes current file written and creates a new one ready to append.
  180. func (w *WAL) Cut() error {
  181. // create a new wal file with name sequence + 1
  182. fpath := path.Join(w.dir, walName(w.seq+1, w.enti+1))
  183. f, err := os.OpenFile(fpath, os.O_WRONLY|os.O_APPEND|os.O_CREATE, 0600)
  184. if err != nil {
  185. return err
  186. }
  187. w.Sync()
  188. w.f.Close()
  189. log.Printf("wal.cut index=%d prevfile=%s curfile=%s", w.enti, w.f.Name(), f.Name())
  190. // update writer and save the previous crc
  191. w.f = f
  192. w.seq++
  193. prevCrc := w.encoder.crc.Sum32()
  194. w.encoder = newEncoder(w.f, prevCrc)
  195. return w.saveCrc(prevCrc)
  196. }
  197. func (w *WAL) Sync() error {
  198. if w.encoder != nil {
  199. if err := w.encoder.flush(); err != nil {
  200. return err
  201. }
  202. }
  203. return w.f.Sync()
  204. }
  205. func (w *WAL) Close() {
  206. log.Printf("path=%s wal.close", w.f.Name())
  207. if w.f != nil {
  208. w.Sync()
  209. w.f.Close()
  210. }
  211. }
  212. func (w *WAL) SaveInfo(i *raftpb.Info) error {
  213. log.Printf("path=%s wal.saveInfo id=%d", w.f.Name(), i.Id)
  214. b, err := i.Marshal()
  215. if err != nil {
  216. panic(err)
  217. }
  218. rec := &walpb.Record{Type: infoType, Data: b}
  219. return w.encoder.encode(rec)
  220. }
  221. func (w *WAL) SaveEntry(e *raftpb.Entry) error {
  222. b, err := e.Marshal()
  223. if err != nil {
  224. panic(err)
  225. }
  226. rec := &walpb.Record{Type: entryType, Data: b}
  227. if err := w.encoder.encode(rec); err != nil {
  228. return err
  229. }
  230. w.enti = e.Index
  231. return nil
  232. }
  233. func (w *WAL) SaveState(s *raftpb.HardState) error {
  234. if raft.IsEmptyHardState(*s) {
  235. return nil
  236. }
  237. log.Printf("path=%s wal.saveState state=\"%+v\"", w.f.Name(), s)
  238. b, err := s.Marshal()
  239. if err != nil {
  240. panic(err)
  241. }
  242. rec := &walpb.Record{Type: stateType, Data: b}
  243. return w.encoder.encode(rec)
  244. }
  245. func (w *WAL) Save(st raftpb.HardState, ents []raftpb.Entry) {
  246. // TODO(xiangli): no more reference operator
  247. w.SaveState(&st)
  248. for i := range ents {
  249. w.SaveEntry(&ents[i])
  250. }
  251. w.Sync()
  252. }
  253. func (w *WAL) saveCrc(prevCrc uint32) error {
  254. return w.encoder.encode(&walpb.Record{Type: crcType, Crc: prevCrc})
  255. }