You can not select more than 25 topics Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.

core.go 12 kB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551
  1. package publock
  2. import (
  3. "fmt"
  4. "io"
  5. "sync"
  6. "time"
  7. "gitlink.org.cn/cloudream/common/pkgs/async"
  8. "gitlink.org.cn/cloudream/common/pkgs/logger"
  9. "gitlink.org.cn/cloudream/common/pkgs/trie"
  10. "gitlink.org.cn/cloudream/common/pkgs/types"
  11. "gitlink.org.cn/cloudream/common/utils/lo2"
  12. "gitlink.org.cn/cloudream/common/utils/serder"
  13. "gitlink.org.cn/cloudream/jcs-pub/client/internal/cluster"
  14. "gitlink.org.cn/cloudream/jcs-pub/client/internal/publock/lockprovider"
  15. pubtypes "gitlink.org.cn/cloudream/jcs-pub/client/internal/publock/types"
  16. "gitlink.org.cn/cloudream/jcs-pub/common/ecode"
  17. )
  18. type Core struct {
  19. cfg Config
  20. clster *cluster.Cluster
  21. fsm *ClusterFSM
  22. lock *sync.Mutex
  23. provdersTrie *trie.Trie[pubtypes.LockProvider]
  24. acquirings []*acquiring // 必须使用数组,因为要保证顺序(集群多个节点的执行结果应该严格相同)
  25. acquireds map[pubtypes.RequestID]*acquired
  26. eventCh *async.UnboundChannel[CoreEvent]
  27. doneCh chan any
  28. }
  29. func NewCore(cfg Config, clster *cluster.Cluster) *Core {
  30. c := &Core{
  31. cfg: cfg,
  32. clster: clster,
  33. lock: &sync.Mutex{},
  34. provdersTrie: trie.NewTrie[pubtypes.LockProvider](),
  35. acquireds: make(map[pubtypes.RequestID]*acquired),
  36. eventCh: async.NewUnboundChannel[CoreEvent](),
  37. doneCh: make(chan any, 1),
  38. }
  39. c.fsm = &ClusterFSM{
  40. core: c,
  41. }
  42. c.provdersTrie.Create([]any{lockprovider.UserSpaceLockPathPrefix, trie.WORD_ANY}).Value = lockprovider.NewUserSpaceLock()
  43. c.provdersTrie.Create([]any{lockprovider.PackageLockPathPrefix, trie.WORD_ANY}).Value = lockprovider.NewPackageLock()
  44. return c
  45. }
  46. func (s *Core) Apply(cmd Command) {
  47. switch cmd := cmd.(type) {
  48. case *Acquire:
  49. s.acquire(*cmd)
  50. case *AcquireTimeout:
  51. s.acquireTimeout(*cmd)
  52. case *Release:
  53. s.release(*cmd)
  54. case *LeaseExpired:
  55. s.leaseExpired(*cmd)
  56. case *Renew:
  57. s.renew(*cmd)
  58. }
  59. }
  60. func (s *Core) EventChan() *async.UnboundChannel[CoreEvent] {
  61. return s.eventCh
  62. }
  63. func (s *Core) FSM() cluster.FSM {
  64. return s.fsm
  65. }
  66. func (s *Core) Start() {
  67. log := logger.WithField("Mod", "Publock.Core")
  68. go func() {
  69. ticker := time.NewTicker(time.Second)
  70. defer ticker.Stop()
  71. loop:
  72. for {
  73. select {
  74. case <-s.doneCh:
  75. break loop
  76. case <-ticker.C:
  77. var acTimeout []pubtypes.RequestID
  78. var leExpired []pubtypes.RequestID
  79. s.lock.Lock()
  80. // 定时清理超时的锁请求
  81. for _, req := range s.acquirings {
  82. if time.Since(req.StartTime) < req.Cmd.Timeout {
  83. continue
  84. }
  85. acTimeout = append(acTimeout, req.Cmd.ID)
  86. }
  87. // 定时清理过期的锁
  88. for reqID, ac := range s.acquireds {
  89. if ac.ExpireCounter < s.cfg.LeaseExpiredSeconds {
  90. ac.ExpireCounter++
  91. continue
  92. }
  93. leExpired = append(leExpired, reqID)
  94. }
  95. s.lock.Unlock()
  96. for _, reqID := range acTimeout {
  97. cmd := AcquireTimeout{
  98. ID: reqID,
  99. }
  100. if s.clster != nil {
  101. data, err := serder.ObjectToJSONEx(cmd)
  102. if err != nil {
  103. log.Warnf("cmd %T to json: %v", cmd, err)
  104. continue
  105. }
  106. // 不管是否成功,有定时任务兜底
  107. s.clster.Apply(s.fsm.ID(), data, time.Second*3)
  108. } else {
  109. s.acquireTimeout(AcquireTimeout{
  110. ID: reqID,
  111. })
  112. }
  113. }
  114. for _, reqID := range leExpired {
  115. cmd := LeaseExpired{
  116. ID: reqID,
  117. }
  118. if s.clster != nil {
  119. data, err := serder.ObjectToJSONEx(cmd)
  120. if err != nil {
  121. log.Warnf("cmd %T to json: %v", cmd, err)
  122. continue
  123. }
  124. // 不管是否成功,有定时任务兜底
  125. s.clster.Apply(s.fsm.ID(), data, time.Second*3)
  126. } else {
  127. s.leaseExpired(LeaseExpired{
  128. ID: reqID,
  129. })
  130. }
  131. }
  132. }
  133. }
  134. }()
  135. }
  136. func (s *Core) Stop() {
  137. select {
  138. case s.doneCh <- nil:
  139. default:
  140. }
  141. s.eventCh.Close()
  142. }
  143. type acquiring struct {
  144. Cmd Acquire
  145. LastErr *ecode.CodeError
  146. // 这个值来自自每个节点自身,所以可能会各不相同。
  147. // 但这个值只是作为判断获取锁是否超时的依据,所以问题不大。
  148. StartTime time.Time
  149. }
  150. type acquired struct {
  151. ID pubtypes.RequestID
  152. Req pubtypes.LockRequest
  153. // 这个值用来记录锁经过的过期检查的次数,超过一定次数则认为过期。
  154. // 因为未通知加锁的服务而释放一个锁是一个危险操作,所以这里采用这种计数的方式来实现过期
  155. ExpireCounter int
  156. }
  157. func (c *Core) acquire(cmd Acquire) {
  158. c.lock.Lock()
  159. defer c.lock.Unlock()
  160. // 立刻检测锁是否可用
  161. cerr := c.tryAcquireOne(cmd.ID, cmd.Request)
  162. if cerr == nil {
  163. c.eventCh.Send(&AcquireResult{
  164. Raw: cmd,
  165. Error: nil,
  166. })
  167. return
  168. }
  169. // 不可用则加入等待列表
  170. info := &acquiring{
  171. Cmd: cmd,
  172. LastErr: cerr,
  173. StartTime: time.Now(),
  174. }
  175. c.acquirings = append(c.acquirings, info)
  176. go func() {
  177. log := logger.WithField("Mod", "Publock.Core")
  178. <-time.After(cmd.Timeout)
  179. ac := AcquireTimeout{
  180. ID: cmd.ID,
  181. }
  182. data, err := serder.ObjectToJSONEx(ac)
  183. if err != nil {
  184. log.Warnf("cmd %T to json: %v", ac, err)
  185. return
  186. }
  187. // 不管是否成功,有定时任务兜底
  188. c.clster.Apply(c.fsm.ID(), data, cmd.Timeout)
  189. }()
  190. }
  191. func (s *Core) release(cmd Release) {
  192. reqID := cmd.ID
  193. s.lock.Lock()
  194. defer s.lock.Unlock()
  195. ac, ok := s.acquireds[reqID]
  196. if !ok {
  197. return
  198. }
  199. s.releaseRequest(reqID, ac.Req)
  200. s.eventCh.Send(&Released{
  201. ID: reqID,
  202. })
  203. s.tryAcquirings()
  204. }
  205. func (c *Core) acquireTimeout(cmd AcquireTimeout) {
  206. c.lock.Lock()
  207. defer c.lock.Unlock()
  208. for i, req := range c.acquirings {
  209. if req.Cmd.ID == cmd.ID {
  210. c.eventCh.Send(&AcquireResult{
  211. Raw: req.Cmd,
  212. Error: req.LastErr,
  213. })
  214. c.acquirings = lo2.RemoveAt(c.acquirings, i)
  215. return
  216. }
  217. }
  218. }
  219. func (c *Core) leaseExpired(cmd LeaseExpired) {
  220. log := logger.WithField("Mod", "Publock.Core")
  221. c.lock.Lock()
  222. defer c.lock.Unlock()
  223. ac, ok := c.acquireds[cmd.ID]
  224. if !ok {
  225. return
  226. }
  227. log.Debugf("lock request %v lease expired", ac.ID)
  228. c.releaseRequest(ac.ID, ac.Req)
  229. c.tryAcquirings()
  230. }
  231. func (c *Core) renew(cmd Renew) {
  232. c.lock.Lock()
  233. defer c.lock.Unlock()
  234. for _, reqID := range cmd.IDs {
  235. ac, ok := c.acquireds[reqID]
  236. if !ok {
  237. continue
  238. }
  239. ac.ExpireCounter = 0
  240. }
  241. }
  242. func (a *Core) tryAcquirings() {
  243. for i := 0; i < len(a.acquirings); i++ {
  244. req := a.acquirings[i]
  245. err := a.tryAcquireOne(req.Cmd.ID, req.Cmd.Request)
  246. if err != nil {
  247. req.LastErr = err
  248. continue
  249. }
  250. a.eventCh.Send(&AcquireResult{
  251. Raw: req.Cmd,
  252. Error: nil,
  253. })
  254. a.acquirings[i] = nil
  255. }
  256. a.acquirings = lo2.RemoveAllDefault(a.acquirings)
  257. }
  258. func (s *Core) tryAcquireOne(reqID pubtypes.RequestID, req pubtypes.LockRequest) *ecode.CodeError {
  259. cerr := s.testOneRequest(req)
  260. if cerr != nil {
  261. return cerr
  262. }
  263. s.applyRequest(reqID, req)
  264. s.acquireds[reqID] = &acquired{
  265. ID: reqID,
  266. Req: req,
  267. ExpireCounter: 0,
  268. }
  269. return nil
  270. }
  271. func (s *Core) testOneRequest(req pubtypes.LockRequest) *ecode.CodeError {
  272. for _, lock := range req.Locks {
  273. n, ok := s.provdersTrie.WalkEnd(lock.Path)
  274. if !ok || n.Value == nil {
  275. return ecode.Newf(ecode.DataNotFound, "lock provider not found for path %v", lock.Path)
  276. }
  277. err := n.Value.CanLock(lock)
  278. if err != nil {
  279. return ecode.Newf(ecode.LockConflict, "%v", err)
  280. }
  281. }
  282. return nil
  283. }
  284. func (s *Core) applyRequest(reqID pubtypes.RequestID, req pubtypes.LockRequest) {
  285. for _, lock := range req.Locks {
  286. p, _ := s.provdersTrie.WalkEnd(lock.Path)
  287. p.Value.Lock(reqID, lock)
  288. }
  289. }
  290. func (s *Core) releaseRequest(reqID pubtypes.RequestID, req pubtypes.LockRequest) {
  291. for _, lock := range req.Locks {
  292. p, _ := s.provdersTrie.WalkEnd(lock.Path)
  293. p.Value.Unlock(reqID, lock)
  294. }
  295. delete(s.acquireds, reqID)
  296. }
  297. type LockedRequest struct {
  298. Req pubtypes.LockRequest
  299. ReqID pubtypes.RequestID
  300. }
  301. type Command interface {
  302. IsCommand() bool
  303. }
  304. var _ = serder.UseTypeUnionExternallyTagged(types.Ref(types.NewTypeUnion[Command](
  305. (*Acquire)(nil),
  306. (*Release)(nil),
  307. (*AcquireTimeout)(nil),
  308. (*LeaseExpired)(nil),
  309. (*Renew)(nil),
  310. )))
  311. type Acquire struct {
  312. ID pubtypes.RequestID
  313. Request pubtypes.LockRequest
  314. Timeout time.Duration
  315. Reason string
  316. }
  317. func (a *Acquire) IsCommand() bool {
  318. return true
  319. }
  320. type Release struct {
  321. ID pubtypes.RequestID
  322. }
  323. func (r *Release) IsCommand() bool {
  324. return true
  325. }
  326. type AcquireTimeout struct {
  327. ID pubtypes.RequestID
  328. }
  329. func (a *AcquireTimeout) IsCommand() bool {
  330. return true
  331. }
  332. type LeaseExpired struct {
  333. ID pubtypes.RequestID
  334. }
  335. func (l *LeaseExpired) IsCommand() bool {
  336. return true
  337. }
  338. type Renew struct {
  339. IDs []pubtypes.RequestID
  340. }
  341. func (r *Renew) IsCommand() bool {
  342. return true
  343. }
  344. type CoreEvent interface {
  345. IsCoreEvent() bool
  346. }
  347. type AcquireResult struct {
  348. Raw Acquire
  349. Error *ecode.CodeError
  350. }
  351. func (a *AcquireResult) IsCoreEvent() bool {
  352. return true
  353. }
  354. type Released struct {
  355. ID pubtypes.RequestID
  356. }
  357. func (r *Released) IsCoreEvent() bool {
  358. return true
  359. }
  360. type ClusterFSM struct {
  361. core *Core
  362. }
  363. func (f *ClusterFSM) ID() string {
  364. return "Publock"
  365. }
  366. func (f *ClusterFSM) Apply(cmdData []byte) ([]byte, error) {
  367. cmd, err := serder.JSONToObjectEx[Command](cmdData)
  368. if err != nil {
  369. return nil, fmt.Errorf("parse command: %v", err)
  370. }
  371. f.core.Apply(cmd)
  372. return nil, nil
  373. }
  374. func (f *ClusterFSM) Snapshot() (cluster.FSMSnapshot, error) {
  375. log := logger.WithField("Mod", "Publock.ClusterFSM")
  376. log.Debugf("make snapshot")
  377. f.core.lock.Lock()
  378. defer f.core.lock.Unlock()
  379. acquireds := make([]*acquired, 0, len(f.core.acquireds))
  380. for _, ac := range f.core.acquireds {
  381. newAc := &acquired{
  382. ID: ac.ID,
  383. Req: ac.Req,
  384. ExpireCounter: ac.ExpireCounter,
  385. }
  386. acquireds = append(acquireds, newAc)
  387. }
  388. acquirings := make([]*acquiring, 0, len(f.core.acquirings))
  389. for _, ac := range f.core.acquirings {
  390. newAc := &acquiring{
  391. Cmd: ac.Cmd,
  392. LastErr: ac.LastErr,
  393. StartTime: ac.StartTime,
  394. }
  395. acquirings = append(acquirings, newAc)
  396. }
  397. return &FSMSnapshot{
  398. Acquireds: acquireds,
  399. Acquirings: acquirings,
  400. }, nil
  401. }
  402. func (f *ClusterFSM) Restore(input io.Reader) error {
  403. log := logger.WithField("Mod", "Publock.ClusterFSM")
  404. log.Debugf("restore from input")
  405. snap := &FSMSnapshot{}
  406. err := serder.JSONToObjectStream(input, snap)
  407. if err != nil {
  408. return err
  409. }
  410. f.core.lock.Lock()
  411. defer f.core.lock.Unlock()
  412. f.core.provdersTrie.Walk(nil, func(word string, wordIndex int, node *trie.Node[pubtypes.LockProvider], isWordNode bool) {
  413. if node.Value != nil {
  414. node.Value.Clear()
  415. }
  416. })
  417. f.core.acquireds = make(map[pubtypes.RequestID]*acquired)
  418. for _, a := range snap.Acquireds {
  419. f.core.applyRequest(a.ID, a.Req)
  420. }
  421. f.core.acquirings = snap.Acquirings
  422. for _, req := range f.core.acquirings {
  423. // 已经超时的请求不启动精确的定时任务
  424. if time.Since(req.StartTime) > req.Cmd.Timeout {
  425. continue
  426. }
  427. go func() {
  428. <-time.After(req.Cmd.Timeout - time.Since(req.StartTime))
  429. cmd := AcquireTimeout{
  430. ID: req.Cmd.ID,
  431. }
  432. data, err := serder.ObjectToJSONEx(cmd)
  433. if err != nil {
  434. log.Warnf("cmd %T to json: %v", cmd, err)
  435. return
  436. }
  437. // 不管是否成功,有定时任务兜底
  438. f.core.clster.Apply(f.core.fsm.ID(), data, req.Cmd.Timeout)
  439. }()
  440. }
  441. return nil
  442. }
  443. type FSMSnapshot struct {
  444. Acquireds []*acquired
  445. Acquirings []*acquiring
  446. }
  447. func (s *FSMSnapshot) Persist(output io.Writer) error {
  448. rc := serder.ObjectToJSONStream(s)
  449. defer rc.Close()
  450. _, err := io.Copy(output, rc)
  451. return err
  452. }
  453. func (s *FSMSnapshot) Release() {}

本项目旨在将云际存储公共基础设施化,使个人及企业可低门槛使用高效的云际存储服务(安装开箱即用云际存储客户端即可,无需关注其他组件的部署),同时支持用户灵活便捷定制云际存储的功能细节。