control.go 8.6 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404
  1. package gocql
  2. import (
  3. crand "crypto/rand"
  4. "errors"
  5. "fmt"
  6. "golang.org/x/net/context"
  7. "log"
  8. "math/rand"
  9. "net"
  10. "strconv"
  11. "sync/atomic"
  12. "time"
  13. )
  14. var (
  15. randr *rand.Rand
  16. )
  17. func init() {
  18. b := make([]byte, 4)
  19. if _, err := crand.Read(b); err != nil {
  20. panic(fmt.Sprintf("unable to seed random number generator: %v", err))
  21. }
  22. randr = rand.New(rand.NewSource(int64(readInt(b))))
  23. }
  24. // Ensure that the atomic variable is aligned to a 64bit boundary
  25. // so that atomic operations can be applied on 32bit architectures.
  26. type controlConn struct {
  27. session *Session
  28. conn atomic.Value
  29. retry RetryPolicy
  30. started int32
  31. quit chan struct{}
  32. }
  33. func createControlConn(session *Session) *controlConn {
  34. control := &controlConn{
  35. session: session,
  36. quit: make(chan struct{}),
  37. retry: &SimpleRetryPolicy{NumRetries: 3},
  38. }
  39. control.conn.Store((*Conn)(nil))
  40. return control
  41. }
  42. func (c *controlConn) heartBeat() {
  43. if !atomic.CompareAndSwapInt32(&c.started, 0, 1) {
  44. return
  45. }
  46. sleepTime := 1 * time.Second
  47. for {
  48. select {
  49. case <-c.quit:
  50. return
  51. case <-time.After(sleepTime):
  52. }
  53. resp, err := c.writeFrame(&writeOptionsFrame{})
  54. if err != nil {
  55. goto reconn
  56. }
  57. switch resp.(type) {
  58. case *supportedFrame:
  59. // Everything ok
  60. sleepTime = 5 * time.Second
  61. continue
  62. case error:
  63. goto reconn
  64. default:
  65. panic(fmt.Sprintf("gocql: unknown frame in response to options: %T", resp))
  66. }
  67. reconn:
  68. // try to connect a bit faster
  69. sleepTime = 1 * time.Second
  70. c.reconnect(true)
  71. // time.Sleep(5 * time.Second)
  72. continue
  73. }
  74. }
  75. func (c *controlConn) shuffleDial(endpoints []string) (conn *Conn, err error) {
  76. perm := randr.Perm(len(endpoints))
  77. shuffled := make([]string, len(endpoints))
  78. for i, endpoint := range endpoints {
  79. shuffled[perm[i]] = endpoint
  80. }
  81. // shuffle endpoints so not all drivers will connect to the same initial
  82. // node.
  83. for _, addr := range shuffled {
  84. if addr == "" {
  85. return nil, fmt.Errorf("control: invalid address: %q", addr)
  86. }
  87. port := c.session.cfg.Port
  88. addr = JoinHostPort(addr, port)
  89. host, portStr, err := net.SplitHostPort(addr)
  90. if err != nil {
  91. host = addr
  92. port = c.session.cfg.Port
  93. err = nil
  94. } else {
  95. port, err = strconv.Atoi(portStr)
  96. if err != nil {
  97. return nil, err
  98. }
  99. }
  100. hostInfo, _ := c.session.ring.addHostIfMissing(&HostInfo{peer: host, port: port})
  101. conn, err = c.session.connect(addr, c, hostInfo)
  102. if err == nil {
  103. return conn, err
  104. }
  105. log.Printf("gocql: unable to dial control conn %v: %v\n", addr, err)
  106. }
  107. return
  108. }
  109. func (c *controlConn) connect(endpoints []string) error {
  110. if len(endpoints) == 0 {
  111. return errors.New("control: no endpoints specified")
  112. }
  113. conn, err := c.shuffleDial(endpoints)
  114. if err != nil {
  115. return fmt.Errorf("control: unable to connect: %v", err)
  116. } else if conn == nil {
  117. return errors.New("control: unable to connect to initial endpoints")
  118. }
  119. if err := c.setupConn(conn); err != nil {
  120. conn.Close()
  121. return fmt.Errorf("control: unable to setup connection: %v", err)
  122. }
  123. // we could fetch the initial ring here and update initial host data. So that
  124. // when we return from here we have a ring topology ready to go.
  125. go c.heartBeat()
  126. return nil
  127. }
  128. func (c *controlConn) setupConn(conn *Conn) error {
  129. if err := c.registerEvents(conn); err != nil {
  130. conn.Close()
  131. return err
  132. }
  133. c.conn.Store(conn)
  134. host, portstr, err := net.SplitHostPort(conn.conn.RemoteAddr().String())
  135. if err != nil {
  136. return err
  137. }
  138. port, err := strconv.Atoi(portstr)
  139. if err != nil {
  140. return err
  141. }
  142. c.session.handleNodeUp(net.ParseIP(host), port, false)
  143. return nil
  144. }
  145. func (c *controlConn) registerEvents(conn *Conn) error {
  146. var events []string
  147. if !c.session.cfg.Events.DisableTopologyEvents {
  148. events = append(events, "TOPOLOGY_CHANGE")
  149. }
  150. if !c.session.cfg.Events.DisableNodeStatusEvents {
  151. events = append(events, "STATUS_CHANGE")
  152. }
  153. if !c.session.cfg.Events.DisableSchemaEvents {
  154. events = append(events, "SCHEMA_CHANGE")
  155. }
  156. if len(events) == 0 {
  157. return nil
  158. }
  159. framer, err := conn.exec(context.Background(),
  160. &writeRegisterFrame{
  161. events: events,
  162. }, nil)
  163. if err != nil {
  164. return err
  165. }
  166. frame, err := framer.parseFrame()
  167. if err != nil {
  168. return err
  169. } else if _, ok := frame.(*readyFrame); !ok {
  170. return fmt.Errorf("unexpected frame in response to register: got %T: %v\n", frame, frame)
  171. }
  172. return nil
  173. }
  174. func (c *controlConn) reconnect(refreshring bool) {
  175. // TODO: simplify this function, use session.ring to get hosts instead of the
  176. // connection pool
  177. addr := c.addr()
  178. oldConn := c.conn.Load().(*Conn)
  179. if oldConn != nil {
  180. oldConn.Close()
  181. }
  182. var newConn *Conn
  183. if addr != "" {
  184. // try to connect to the old host
  185. conn, err := c.session.connect(addr, c, oldConn.host)
  186. if err != nil {
  187. // host is dead
  188. // TODO: this is replicated in a few places
  189. ip, portStr, _ := net.SplitHostPort(addr)
  190. port, _ := strconv.Atoi(portStr)
  191. c.session.handleNodeDown(net.ParseIP(ip), port)
  192. } else {
  193. newConn = conn
  194. }
  195. }
  196. // TODO: should have our own roundrobbin for hosts so that we can try each
  197. // in succession and guantee that we get a different host each time.
  198. if newConn == nil {
  199. host := c.session.ring.rrHost()
  200. if host == nil {
  201. c.connect(c.session.ring.endpoints)
  202. return
  203. }
  204. var err error
  205. newConn, err = c.session.connect(host.Peer(), c, host)
  206. if err != nil {
  207. // TODO: add log handler for things like this
  208. return
  209. }
  210. }
  211. if err := c.setupConn(newConn); err != nil {
  212. newConn.Close()
  213. log.Printf("gocql: control unable to register events: %v\n", err)
  214. return
  215. }
  216. if refreshring {
  217. c.session.hostSource.refreshRing()
  218. }
  219. }
  220. func (c *controlConn) HandleError(conn *Conn, err error, closed bool) {
  221. if !closed {
  222. return
  223. }
  224. oldConn := c.conn.Load().(*Conn)
  225. if oldConn != conn {
  226. return
  227. }
  228. c.reconnect(true)
  229. }
  230. func (c *controlConn) writeFrame(w frameWriter) (frame, error) {
  231. conn := c.conn.Load().(*Conn)
  232. if conn == nil {
  233. return nil, errNoControl
  234. }
  235. framer, err := conn.exec(context.Background(), w, nil)
  236. if err != nil {
  237. return nil, err
  238. }
  239. return framer.parseFrame()
  240. }
  241. func (c *controlConn) withConn(fn func(*Conn) *Iter) *Iter {
  242. const maxConnectAttempts = 5
  243. connectAttempts := 0
  244. for i := 0; i < maxConnectAttempts; i++ {
  245. conn := c.conn.Load().(*Conn)
  246. if conn == nil {
  247. if connectAttempts > maxConnectAttempts {
  248. break
  249. }
  250. connectAttempts++
  251. c.reconnect(false)
  252. continue
  253. }
  254. return fn(conn)
  255. }
  256. return &Iter{err: errNoControl}
  257. }
  258. // query will return nil if the connection is closed or nil
  259. func (c *controlConn) query(statement string, values ...interface{}) (iter *Iter) {
  260. q := c.session.Query(statement, values...).Consistency(One).RoutingKey([]byte{})
  261. for {
  262. iter = c.withConn(func(conn *Conn) *Iter {
  263. return conn.executeQuery(q)
  264. })
  265. if gocqlDebug && iter.err != nil {
  266. log.Printf("control: error executing %q: %v\n", statement, iter.err)
  267. }
  268. q.attempts++
  269. if iter.err == nil || !c.retry.Attempt(q) {
  270. break
  271. }
  272. }
  273. return
  274. }
  275. func (c *controlConn) fetchHostInfo(addr net.IP, port int) (*HostInfo, error) {
  276. // TODO(zariel): we should probably move this into host_source or atleast
  277. // share code with it.
  278. hostname, _, err := net.SplitHostPort(c.addr())
  279. if err != nil {
  280. return nil, fmt.Errorf("unable to fetch host info, invalid conn addr: %q: %v", c.addr(), err)
  281. }
  282. isLocal := hostname == addr.String()
  283. var fn func(*HostInfo) error
  284. if isLocal {
  285. fn = func(host *HostInfo) error {
  286. // TODO(zariel): should we fetch rpc_address from here?
  287. iter := c.query("SELECT data_center, rack, host_id, tokens, release_version FROM system.local WHERE key='local'")
  288. iter.Scan(&host.dataCenter, &host.rack, &host.hostId, &host.tokens, &host.version)
  289. return iter.Close()
  290. }
  291. } else {
  292. fn = func(host *HostInfo) error {
  293. // TODO(zariel): should we fetch rpc_address from here?
  294. iter := c.query("SELECT data_center, rack, host_id, tokens, release_version FROM system.peers WHERE peer=?", addr)
  295. iter.Scan(&host.dataCenter, &host.rack, &host.hostId, &host.tokens, &host.version)
  296. return iter.Close()
  297. }
  298. }
  299. host := &HostInfo{
  300. port: port,
  301. }
  302. if err := fn(host); err != nil {
  303. return nil, err
  304. }
  305. host.peer = addr.String()
  306. return host, nil
  307. }
  308. func (c *controlConn) awaitSchemaAgreement() error {
  309. return c.withConn(func(conn *Conn) *Iter {
  310. return &Iter{err: conn.awaitSchemaAgreement()}
  311. }).err
  312. }
  313. func (c *controlConn) addr() string {
  314. conn := c.conn.Load().(*Conn)
  315. if conn == nil {
  316. return ""
  317. }
  318. return conn.addr
  319. }
  320. func (c *controlConn) close() {
  321. if atomic.CompareAndSwapInt32(&c.started, 1, -1) {
  322. c.quit <- struct{}{}
  323. }
  324. conn := c.conn.Load().(*Conn)
  325. if conn != nil {
  326. conn.Close()
  327. }
  328. }
  329. var errNoControl = errors.New("gocql: no control connection available")