conn.go 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647
  1. // Copyright (c) 2012 The gocql Authors. All rights reserved.
  2. // Use of this source code is governed by a BSD-style
  3. // license that can be found in the LICENSE file.
  4. package gocql
  5. import (
  6. "bufio"
  7. "fmt"
  8. "net"
  9. "strings"
  10. "sync"
  11. "sync/atomic"
  12. "time"
  13. "unicode"
  14. "code.google.com/p/snappy-go/snappy"
  15. )
  16. const defaultFrameSize = 4096
  17. const flagResponse = 0x80
  18. const maskVersion = 0x7F
  19. type Cluster interface {
  20. HandleError(conn *Conn, err error, closed bool)
  21. HandleKeyspace(conn *Conn, keyspace string)
  22. }
  23. type Authenticator interface {
  24. Challenge(req []byte) (resp []byte, auth Authenticator, err error)
  25. Success(data []byte) error
  26. }
  27. type PasswordAuthenticator struct {
  28. Username string
  29. Password string
  30. }
  31. func (p PasswordAuthenticator) Challenge(req []byte) ([]byte, Authenticator, error) {
  32. if string(req) != "org.apache.cassandra.auth.PasswordAuthenticator" {
  33. return nil, nil, fmt.Errorf("unexpected authenticator %q", req)
  34. }
  35. resp := make([]byte, 2+len(p.Username)+len(p.Password))
  36. resp[0] = 0
  37. copy(resp[1:], p.Username)
  38. resp[len(p.Username)+1] = 0
  39. copy(resp[2+len(p.Username):], p.Password)
  40. return resp, nil, nil
  41. }
  42. func (p PasswordAuthenticator) Success(data []byte) error {
  43. return nil
  44. }
  45. type ConnConfig struct {
  46. ProtoVersion int
  47. CQLVersion string
  48. Timeout time.Duration
  49. NumStreams int
  50. Compressor Compressor
  51. Authenticator Authenticator
  52. Keepalive time.Duration
  53. }
  54. // Conn is a single connection to a Cassandra node. It can be used to execute
  55. // queries, but users are usually advised to use a more reliable, higher
  56. // level API.
  57. type Conn struct {
  58. conn net.Conn
  59. r *bufio.Reader
  60. timeout time.Duration
  61. uniq chan uint8
  62. calls []callReq
  63. nwait int32
  64. prepMu sync.Mutex
  65. prep map[string]*inflightPrepare
  66. cluster Cluster
  67. compressor Compressor
  68. auth Authenticator
  69. addr string
  70. version uint8
  71. }
  72. // Connect establishes a connection to a Cassandra node.
  73. // You must also call the Serve method before you can execute any queries.
  74. func Connect(addr string, cfg ConnConfig, cluster Cluster) (*Conn, error) {
  75. conn, err := net.DialTimeout("tcp", addr, cfg.Timeout)
  76. if err != nil {
  77. return nil, err
  78. }
  79. if cfg.NumStreams <= 0 || cfg.NumStreams > 128 {
  80. cfg.NumStreams = 128
  81. }
  82. if cfg.ProtoVersion != 1 && cfg.ProtoVersion != 2 {
  83. cfg.ProtoVersion = 2
  84. }
  85. c := &Conn{
  86. conn: conn,
  87. r: bufio.NewReader(conn),
  88. uniq: make(chan uint8, cfg.NumStreams),
  89. calls: make([]callReq, cfg.NumStreams),
  90. prep: make(map[string]*inflightPrepare),
  91. timeout: cfg.Timeout,
  92. version: uint8(cfg.ProtoVersion),
  93. addr: conn.RemoteAddr().String(),
  94. cluster: cluster,
  95. compressor: cfg.Compressor,
  96. auth: cfg.Authenticator,
  97. }
  98. if cfg.Keepalive > 0 {
  99. c.setKeepalive(cfg.Keepalive)
  100. }
  101. for i := 0; i < cap(c.uniq); i++ {
  102. c.uniq <- uint8(i)
  103. }
  104. if err := c.startup(&cfg); err != nil {
  105. return nil, err
  106. }
  107. go c.serve()
  108. return c, nil
  109. }
  110. func (c *Conn) startup(cfg *ConnConfig) error {
  111. compression := ""
  112. if c.compressor != nil {
  113. compression = c.compressor.Name()
  114. }
  115. var req operation = &startupFrame{
  116. CQLVersion: cfg.CQLVersion,
  117. Compression: compression,
  118. }
  119. var challenger Authenticator
  120. for {
  121. resp, err := c.execSimple(req)
  122. if err != nil {
  123. return err
  124. }
  125. switch x := resp.(type) {
  126. case readyFrame:
  127. return nil
  128. case error:
  129. return x
  130. case authenticateFrame:
  131. if c.auth == nil {
  132. return fmt.Errorf("authentication required (using %q)", x.Authenticator)
  133. }
  134. var resp []byte
  135. resp, challenger, err = c.auth.Challenge([]byte(x.Authenticator))
  136. if err != nil {
  137. return err
  138. }
  139. req = &authResponseFrame{resp}
  140. case authChallengeFrame:
  141. if challenger == nil {
  142. return fmt.Errorf("authentication error (invalid challenge)")
  143. }
  144. var resp []byte
  145. resp, challenger, err = challenger.Challenge(x.Data)
  146. if err != nil {
  147. return err
  148. }
  149. req = &authResponseFrame{resp}
  150. case authSuccessFrame:
  151. if challenger != nil {
  152. return challenger.Success(x.Data)
  153. }
  154. return nil
  155. default:
  156. return ErrProtocol
  157. }
  158. }
  159. }
  160. // Serve starts the stream multiplexer for this connection, which is required
  161. // to execute any queries. This method runs as long as the connection is
  162. // open and is therefore usually called in a separate goroutine.
  163. func (c *Conn) serve() {
  164. var (
  165. err error
  166. resp frame
  167. )
  168. for {
  169. resp, err = c.recv()
  170. if err != nil {
  171. break
  172. }
  173. c.dispatch(resp)
  174. }
  175. c.conn.Close()
  176. for id := 0; id < len(c.calls); id++ {
  177. req := &c.calls[id]
  178. if atomic.LoadInt32(&req.active) == 1 {
  179. req.resp <- callResp{nil, err}
  180. }
  181. }
  182. c.cluster.HandleError(c, err, true)
  183. }
  184. func (c *Conn) recv() (frame, error) {
  185. resp := make(frame, headerSize, headerSize+512)
  186. c.conn.SetReadDeadline(time.Now().Add(c.timeout))
  187. n, last, pinged := 0, 0, false
  188. for n < len(resp) {
  189. nn, err := c.r.Read(resp[n:])
  190. n += nn
  191. if err != nil {
  192. if nerr, ok := err.(net.Error); ok && nerr.Timeout() {
  193. if n > last {
  194. // we hit the deadline but we made progress.
  195. // simply extend the deadline
  196. c.conn.SetReadDeadline(time.Now().Add(c.timeout))
  197. last = n
  198. } else if n == 0 && !pinged {
  199. c.conn.SetReadDeadline(time.Now().Add(c.timeout))
  200. if atomic.LoadInt32(&c.nwait) > 0 {
  201. go c.ping()
  202. pinged = true
  203. }
  204. } else {
  205. return nil, err
  206. }
  207. } else {
  208. return nil, err
  209. }
  210. }
  211. if n == headerSize && len(resp) == headerSize {
  212. if resp[0] != c.version|flagResponse {
  213. return nil, ErrProtocol
  214. }
  215. resp.grow(resp.Length())
  216. }
  217. }
  218. return resp, nil
  219. }
  220. func (c *Conn) execSimple(op operation) (interface{}, error) {
  221. f, err := op.encodeFrame(c.version, nil)
  222. f.setLength(len(f) - headerSize)
  223. if _, err := c.conn.Write([]byte(f)); err != nil {
  224. c.conn.Close()
  225. return nil, err
  226. }
  227. if f, err = c.recv(); err != nil {
  228. return nil, err
  229. }
  230. return c.decodeFrame(f, nil)
  231. }
  232. func (c *Conn) exec(op operation, trace Tracer) (interface{}, error) {
  233. req, err := op.encodeFrame(c.version, nil)
  234. if err != nil {
  235. return nil, err
  236. }
  237. if trace != nil {
  238. req[1] |= flagTrace
  239. }
  240. if len(req) > headerSize && c.compressor != nil {
  241. body, err := c.compressor.Encode([]byte(req[headerSize:]))
  242. if err != nil {
  243. return nil, err
  244. }
  245. req = append(req[:headerSize], frame(body)...)
  246. req[1] |= flagCompress
  247. }
  248. req.setLength(len(req) - headerSize)
  249. id := <-c.uniq
  250. req[2] = id
  251. call := &c.calls[id]
  252. call.resp = make(chan callResp, 1)
  253. atomic.AddInt32(&c.nwait, 1)
  254. atomic.StoreInt32(&call.active, 1)
  255. if n, err := c.conn.Write(req); err != nil {
  256. c.conn.Close()
  257. c.uniq <- id
  258. if n > 0 {
  259. return nil, ErrProtocol
  260. }
  261. return nil, ErrUnavailable
  262. }
  263. reply := <-call.resp
  264. call.resp = nil
  265. c.uniq <- id
  266. if reply.err != nil {
  267. return nil, reply.err
  268. }
  269. return c.decodeFrame(reply.buf, trace)
  270. }
  271. func (c *Conn) dispatch(resp frame) {
  272. id := int(resp[2])
  273. if id >= len(c.calls) {
  274. return
  275. }
  276. call := &c.calls[id]
  277. if !atomic.CompareAndSwapInt32(&call.active, 1, 0) {
  278. return
  279. }
  280. atomic.AddInt32(&c.nwait, -1)
  281. call.resp <- callResp{resp, nil}
  282. }
  283. func (c *Conn) ping() error {
  284. _, err := c.exec(&optionsFrame{}, nil)
  285. return err
  286. }
  287. func (c *Conn) prepareStatement(stmt string, trace Tracer) (*queryInfo, error) {
  288. c.prepMu.Lock()
  289. flight := c.prep[stmt]
  290. if flight != nil {
  291. c.prepMu.Unlock()
  292. flight.wg.Wait()
  293. return flight.info, flight.err
  294. }
  295. flight = new(inflightPrepare)
  296. flight.wg.Add(1)
  297. c.prep[stmt] = flight
  298. c.prepMu.Unlock()
  299. resp, err := c.exec(&prepareFrame{Stmt: stmt}, trace)
  300. if err != nil {
  301. flight.err = err
  302. } else {
  303. switch x := resp.(type) {
  304. case resultPreparedFrame:
  305. flight.info = &queryInfo{
  306. id: x.PreparedId,
  307. args: x.Values,
  308. }
  309. case error:
  310. flight.err = x
  311. default:
  312. flight.err = ErrProtocol
  313. }
  314. }
  315. flight.wg.Done()
  316. if err != nil {
  317. c.prepMu.Lock()
  318. delete(c.prep, stmt)
  319. c.prepMu.Unlock()
  320. }
  321. return flight.info, flight.err
  322. }
  323. func shouldPrepare(stmt string) bool {
  324. stmt = strings.TrimLeftFunc(strings.TrimRightFunc(stmt, func(r rune) bool {
  325. return unicode.IsSpace(r) || r == ';'
  326. }), unicode.IsSpace)
  327. var stmtType string
  328. if n := strings.IndexFunc(stmt, unicode.IsSpace); n >= 0 {
  329. stmtType = strings.ToLower(stmt[:n])
  330. }
  331. if stmtType == "begin" {
  332. if n := strings.LastIndexFunc(stmt, unicode.IsSpace); n >= 0 {
  333. stmtType = strings.ToLower(stmt[n+1:])
  334. }
  335. }
  336. switch stmtType {
  337. case "select", "insert", "update", "delete", "batch":
  338. return true
  339. }
  340. return false
  341. }
  342. func (c *Conn) executeQuery(qry *Query) *Iter {
  343. op := &queryFrame{
  344. Stmt: qry.stmt,
  345. Cons: qry.cons,
  346. PageSize: qry.pageSize,
  347. PageState: qry.pageState,
  348. }
  349. if shouldPrepare(op.Stmt) {
  350. // Prepare all DML queries. Other queries can not be prepared.
  351. info, err := c.prepareStatement(qry.stmt, qry.trace)
  352. if err != nil {
  353. return &Iter{err: err}
  354. }
  355. op.Prepared = info.id
  356. op.Values = make([][]byte, len(qry.values))
  357. for i := 0; i < len(qry.values); i++ {
  358. val, err := Marshal(info.args[i].TypeInfo, qry.values[i])
  359. if err != nil {
  360. return &Iter{err: err}
  361. }
  362. op.Values[i] = val
  363. }
  364. }
  365. resp, err := c.exec(op, qry.trace)
  366. if err != nil {
  367. return &Iter{err: err}
  368. }
  369. switch x := resp.(type) {
  370. case resultVoidFrame:
  371. return &Iter{}
  372. case resultRowsFrame:
  373. iter := &Iter{columns: x.Columns, rows: x.Rows}
  374. if len(x.PagingState) > 0 {
  375. iter.next = &nextIter{
  376. qry: *qry,
  377. pos: int((1 - qry.prefetch) * float64(len(iter.rows))),
  378. }
  379. iter.next.qry.pageState = x.PagingState
  380. if iter.next.pos < 1 {
  381. iter.next.pos = 1
  382. }
  383. }
  384. return iter
  385. case resultKeyspaceFrame:
  386. c.cluster.HandleKeyspace(c, x.Keyspace)
  387. return &Iter{}
  388. case errorFrame:
  389. if x.Code == errUnprepared && len(qry.values) > 0 {
  390. c.prepMu.Lock()
  391. if val, ok := c.prep[qry.stmt]; ok && val != nil {
  392. delete(c.prep, qry.stmt)
  393. c.prepMu.Unlock()
  394. return c.executeQuery(qry)
  395. }
  396. c.prepMu.Unlock()
  397. return &Iter{err: x}
  398. } else {
  399. return &Iter{err: x}
  400. }
  401. case error:
  402. return &Iter{err: x}
  403. default:
  404. return &Iter{err: ErrProtocol}
  405. }
  406. }
  407. func (c *Conn) Pick(qry *Query) *Conn {
  408. return c
  409. }
  410. func (c *Conn) Close() {
  411. c.conn.Close()
  412. }
  413. func (c *Conn) Address() string {
  414. return c.addr
  415. }
  416. func (c *Conn) UseKeyspace(keyspace string) error {
  417. resp, err := c.exec(&queryFrame{Stmt: `USE "` + keyspace + `"`, Cons: Any}, nil)
  418. if err != nil {
  419. return err
  420. }
  421. switch x := resp.(type) {
  422. case resultKeyspaceFrame:
  423. case error:
  424. return x
  425. default:
  426. return ErrProtocol
  427. }
  428. return nil
  429. }
  430. func (c *Conn) executeBatch(batch *Batch) error {
  431. if c.version == 1 {
  432. return ErrUnsupported
  433. }
  434. f := make(frame, headerSize, defaultFrameSize)
  435. f.setHeader(c.version, 0, 0, opBatch)
  436. f.writeByte(byte(batch.Type))
  437. f.writeShort(uint16(len(batch.Entries)))
  438. for i := 0; i < len(batch.Entries); i++ {
  439. entry := &batch.Entries[i]
  440. var info *queryInfo
  441. if len(entry.Args) > 0 {
  442. var err error
  443. info, err = c.prepareStatement(entry.Stmt, nil)
  444. if err != nil {
  445. return err
  446. }
  447. f.writeByte(1)
  448. f.writeShortBytes(info.id)
  449. } else {
  450. f.writeByte(0)
  451. f.writeLongString(entry.Stmt)
  452. }
  453. f.writeShort(uint16(len(entry.Args)))
  454. for j := 0; j < len(entry.Args); j++ {
  455. val, err := Marshal(info.args[j].TypeInfo, entry.Args[j])
  456. if err != nil {
  457. return err
  458. }
  459. f.writeBytes(val)
  460. }
  461. }
  462. f.writeConsistency(batch.Cons)
  463. resp, err := c.exec(f, nil)
  464. if err != nil {
  465. return err
  466. }
  467. switch x := resp.(type) {
  468. case resultVoidFrame:
  469. return nil
  470. case error:
  471. return x
  472. default:
  473. return ErrProtocol
  474. }
  475. }
  476. func (c *Conn) decodeFrame(f frame, trace Tracer) (rval interface{}, err error) {
  477. defer func() {
  478. if r := recover(); r != nil {
  479. if e, ok := r.(error); ok && e == ErrProtocol {
  480. err = e
  481. return
  482. }
  483. panic(r)
  484. }
  485. }()
  486. if len(f) < headerSize || (f[0] != c.version|flagResponse) {
  487. return nil, ErrProtocol
  488. }
  489. flags, op, f := f[1], f[3], f[headerSize:]
  490. if flags&flagCompress != 0 && len(f) > 0 && c.compressor != nil {
  491. if buf, err := c.compressor.Decode([]byte(f)); err != nil {
  492. return nil, err
  493. } else {
  494. f = frame(buf)
  495. }
  496. }
  497. if flags&flagTrace != 0 {
  498. if len(f) < 16 {
  499. return nil, ErrProtocol
  500. }
  501. traceId := []byte(f[:16])
  502. f = f[16:]
  503. trace.Trace(traceId)
  504. }
  505. switch op {
  506. case opReady:
  507. return readyFrame{}, nil
  508. case opResult:
  509. switch kind := f.readInt(); kind {
  510. case resultKindVoid:
  511. return resultVoidFrame{}, nil
  512. case resultKindRows:
  513. columns, pageState := f.readMetaData()
  514. numRows := f.readInt()
  515. values := make([][]byte, numRows*len(columns))
  516. for i := 0; i < len(values); i++ {
  517. values[i] = f.readBytes()
  518. }
  519. rows := make([][][]byte, numRows)
  520. for i := 0; i < numRows; i++ {
  521. rows[i], values = values[:len(columns)], values[len(columns):]
  522. }
  523. return resultRowsFrame{columns, rows, pageState}, nil
  524. case resultKindKeyspace:
  525. keyspace := f.readString()
  526. return resultKeyspaceFrame{keyspace}, nil
  527. case resultKindPrepared:
  528. id := f.readShortBytes()
  529. values, _ := f.readMetaData()
  530. return resultPreparedFrame{id, values}, nil
  531. case resultKindSchemaChanged:
  532. return resultVoidFrame{}, nil
  533. default:
  534. return nil, ErrProtocol
  535. }
  536. case opAuthenticate:
  537. return authenticateFrame{f.readString()}, nil
  538. case opAuthChallenge:
  539. return authChallengeFrame{f.readBytes()}, nil
  540. case opAuthSuccess:
  541. return authSuccessFrame{f.readBytes()}, nil
  542. case opSupported:
  543. return supportedFrame{}, nil
  544. case opError:
  545. code := f.readInt()
  546. msg := f.readString()
  547. return errorFrame{code, msg}, nil
  548. default:
  549. return nil, ErrProtocol
  550. }
  551. }
  552. type queryInfo struct {
  553. id []byte
  554. args []ColumnInfo
  555. rval []ColumnInfo
  556. }
  557. type callReq struct {
  558. active int32
  559. resp chan callResp
  560. }
  561. type callResp struct {
  562. buf frame
  563. err error
  564. }
  565. type Compressor interface {
  566. Name() string
  567. Encode(data []byte) ([]byte, error)
  568. Decode(data []byte) ([]byte, error)
  569. }
  570. type inflightPrepare struct {
  571. info *queryInfo
  572. err error
  573. wg sync.WaitGroup
  574. }
  575. // SnappyCompressor implements the Compressor interface and can be used to
  576. // compress incoming and outgoing frames. The snappy compression algorithm
  577. // aims for very high speeds and reasonable compression.
  578. type SnappyCompressor struct{}
  579. func (s SnappyCompressor) Name() string {
  580. return "snappy"
  581. }
  582. func (s SnappyCompressor) Encode(data []byte) ([]byte, error) {
  583. return snappy.Encode(nil, data)
  584. }
  585. func (s SnappyCompressor) Decode(data []byte) ([]byte, error) {
  586. return snappy.Decode(nil, data)
  587. }