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

180 lines
4.5 KiB

11 years ago
11 years ago
11 years ago
  1. package weed_server
  2. import (
  3. "bytes"
  4. "github.com/chrislusf/weed-fs/go/glog"
  5. "github.com/chrislusf/weed-fs/go/topology"
  6. "encoding/json"
  7. "errors"
  8. "fmt"
  9. "github.com/goraft/raft"
  10. "github.com/gorilla/mux"
  11. "io/ioutil"
  12. "math/rand"
  13. "net/http"
  14. "net/url"
  15. "strings"
  16. "time"
  17. )
  18. type RaftServer struct {
  19. peers []string // initial peers to join with
  20. raftServer raft.Server
  21. dataDir string
  22. httpAddr string
  23. router *mux.Router
  24. topo *topology.Topology
  25. }
  26. func NewRaftServer(r *mux.Router, peers []string, httpAddr string, dataDir string, topo *topology.Topology, pulseSeconds int) *RaftServer {
  27. s := &RaftServer{
  28. peers: peers,
  29. httpAddr: httpAddr,
  30. dataDir: dataDir,
  31. router: r,
  32. topo: topo,
  33. }
  34. if glog.V(4) {
  35. raft.SetLogLevel(2)
  36. }
  37. raft.RegisterCommand(&topology.MaxVolumeIdCommand{})
  38. var err error
  39. transporter := raft.NewHTTPTransporter("/cluster", 0)
  40. transporter.Transport.MaxIdleConnsPerHost = 1024
  41. glog.V(1).Infof("Starting RaftServer with IP:%v:", httpAddr)
  42. s.raftServer, err = raft.NewServer(s.httpAddr, s.dataDir, transporter, nil, topo, "")
  43. if err != nil {
  44. glog.V(0).Infoln(err)
  45. return nil
  46. }
  47. transporter.Install(s.raftServer, s)
  48. s.raftServer.SetHeartbeatInterval(1 * time.Second)
  49. s.raftServer.SetElectionTimeout(time.Duration(pulseSeconds) * 1150 * time.Millisecond)
  50. s.raftServer.Start()
  51. s.router.HandleFunc("/cluster/join", s.joinHandler).Methods("POST")
  52. s.router.HandleFunc("/cluster/status", s.statusHandler).Methods("GET")
  53. // Join to leader if specified.
  54. if len(s.peers) > 0 {
  55. if !s.raftServer.IsLogEmpty() {
  56. glog.V(0).Infoln("Starting cluster with existing logs.")
  57. } else {
  58. glog.V(0).Infoln("Joining cluster:", strings.Join(s.peers, ","))
  59. time.Sleep(time.Duration(rand.Intn(1000)) * time.Millisecond)
  60. firstJoinError := s.Join(s.peers)
  61. if firstJoinError != nil {
  62. glog.V(0).Infoln("No existing server found. Starting as leader in the new cluster.")
  63. _, err := s.raftServer.Do(&raft.DefaultJoinCommand{
  64. Name: s.raftServer.Name(),
  65. ConnectionString: "http://" + s.httpAddr,
  66. })
  67. if err != nil {
  68. glog.V(0).Infoln(err)
  69. return nil
  70. }
  71. }
  72. }
  73. // Initialize the server by joining itself.
  74. } else if s.raftServer.IsLogEmpty() {
  75. glog.V(0).Infoln("Initializing new cluster")
  76. _, err := s.raftServer.Do(&raft.DefaultJoinCommand{
  77. Name: s.raftServer.Name(),
  78. ConnectionString: "http://" + s.httpAddr,
  79. })
  80. if err != nil {
  81. glog.V(0).Infoln(err)
  82. return nil
  83. }
  84. } else {
  85. glog.V(0).Infoln("Recovered from log")
  86. }
  87. return s
  88. }
  89. func (s *RaftServer) Peers() (members []string) {
  90. peers := s.raftServer.Peers()
  91. for _, p := range peers {
  92. members = append(members, strings.TrimPrefix(p.ConnectionString, "http://"))
  93. }
  94. return
  95. }
  96. // Join joins an existing cluster.
  97. func (s *RaftServer) Join(peers []string) error {
  98. command := &raft.DefaultJoinCommand{
  99. Name: s.raftServer.Name(),
  100. ConnectionString: "http://" + s.httpAddr,
  101. }
  102. var err error
  103. var b bytes.Buffer
  104. json.NewEncoder(&b).Encode(command)
  105. for _, m := range peers {
  106. if m == s.httpAddr {
  107. continue
  108. }
  109. target := fmt.Sprintf("http://%s/cluster/join", strings.TrimSpace(m))
  110. glog.V(0).Infoln("Attempting to connect to:", target)
  111. err = postFollowingOneRedirect(target, "application/json", &b)
  112. if err != nil {
  113. glog.V(0).Infoln("Post returned error: ", err.Error())
  114. if _, ok := err.(*url.Error); ok {
  115. // If we receive a network error try the next member
  116. continue
  117. }
  118. } else {
  119. return nil
  120. }
  121. }
  122. return errors.New("Could not connect to any cluster peers")
  123. }
  124. // a workaround because http POST following redirection misses request body
  125. func postFollowingOneRedirect(target string, contentType string, b *bytes.Buffer) error {
  126. backupReader := bytes.NewReader(b.Bytes())
  127. resp, err := http.Post(target, contentType, b)
  128. if err != nil {
  129. return err
  130. }
  131. defer resp.Body.Close()
  132. reply, _ := ioutil.ReadAll(resp.Body)
  133. statusCode := resp.StatusCode
  134. if statusCode == http.StatusMovedPermanently {
  135. var urlStr string
  136. if urlStr = resp.Header.Get("Location"); urlStr == "" {
  137. return errors.New(fmt.Sprintf("%d response missing Location header", resp.StatusCode))
  138. }
  139. glog.V(0).Infoln("Post redirected to ", urlStr)
  140. resp2, err2 := http.Post(urlStr, contentType, backupReader)
  141. if err2 != nil {
  142. return err2
  143. }
  144. defer resp2.Body.Close()
  145. reply, _ = ioutil.ReadAll(resp2.Body)
  146. statusCode = resp2.StatusCode
  147. }
  148. glog.V(0).Infoln("Post returned status: ", statusCode, string(reply))
  149. if statusCode != http.StatusOK {
  150. return errors.New(string(reply))
  151. }
  152. return nil
  153. }