mq.go 8.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272
  1. package mq
  2. import (
  3. "context"
  4. "encoding/json"
  5. "errors"
  6. "fmt"
  7. "log"
  8. "strings"
  9. "time"
  10. mqtt "github.com/eclipse/paho.mqtt.golang"
  11. "github.com/gravitl/netmaker/database"
  12. "github.com/gravitl/netmaker/logger"
  13. "github.com/gravitl/netmaker/logic"
  14. "github.com/gravitl/netmaker/models"
  15. "github.com/gravitl/netmaker/netclient/ncutils"
  16. "github.com/gravitl/netmaker/servercfg"
  17. "github.com/gravitl/netmaker/serverctl"
  18. )
  19. const KEEPALIVE_TIMEOUT = 60 //timeout in seconds
  20. const MQ_DISCONNECT = 250
  21. // DefaultHandler default message queue handler - only called when GetDebug == true
  22. var DefaultHandler mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  23. logger.Log(0, "MQTT Message: Topic: ", string(msg.Topic()), " Message: ", string(msg.Payload()))
  24. }
  25. // Ping message Handler -- handles ping topic from client nodes
  26. var Ping mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  27. logger.Log(0, "Ping Handler: ", msg.Topic())
  28. go func() {
  29. id, err := GetID(msg.Topic())
  30. if err != nil {
  31. logger.Log(0, "error getting node.ID sent on ping topic ")
  32. return
  33. }
  34. node, err := logic.GetNodeByID(id)
  35. if err != nil {
  36. logger.Log(0, "mq-ping error getting node: ", err.Error())
  37. record, err := database.FetchRecord(database.NODES_TABLE_NAME, id)
  38. if err != nil {
  39. logger.Log(0, "error reading database ", err.Error())
  40. return
  41. }
  42. logger.Log(0, "record from database")
  43. logger.Log(0, record)
  44. return
  45. }
  46. _, decryptErr := decryptMsg(&node, msg.Payload())
  47. if decryptErr != nil {
  48. logger.Log(0, "error updating node ", node.ID, err.Error())
  49. return
  50. }
  51. node.SetLastCheckIn()
  52. if err := logic.UpdateNode(&node, &node); err != nil {
  53. logger.Log(0, "error updating node ", err.Error())
  54. }
  55. logger.Log(3, "ping processed for node", node.ID)
  56. // --TODO --set client version once feature is implemented.
  57. //node.SetClientVersion(msg.Payload())
  58. }()
  59. }
  60. // UpdateNode message Handler -- handles updates from client nodes
  61. var UpdateNode mqtt.MessageHandler = func(client mqtt.Client, msg mqtt.Message) {
  62. go func() {
  63. id, err := GetID(msg.Topic())
  64. if err != nil {
  65. logger.Log(1, "error getting node.ID sent on ", msg.Topic(), err.Error())
  66. return
  67. }
  68. currentNode, err := logic.GetNodeByID(id)
  69. if err != nil {
  70. logger.Log(1, "error getting node ", id, err.Error())
  71. return
  72. }
  73. decrypted, decryptErr := decryptMsg(&currentNode, msg.Payload())
  74. if decryptErr != nil {
  75. logger.Log(1, "failed to decrypt message for node ", id, decryptErr.Error())
  76. return
  77. }
  78. logger.Log(1, "Update Node Handler", id)
  79. var newNode models.Node
  80. if err := json.Unmarshal(decrypted, &newNode); err != nil {
  81. logger.Log(1, "error unmarshaling payload ", err.Error())
  82. return
  83. }
  84. if err := logic.UpdateNode(&currentNode, &newNode); err != nil {
  85. logger.Log(1, "error saving node", err.Error())
  86. }
  87. if err := PublishPeerUpdate(&newNode); err != nil {
  88. logger.Log(1, "error publishing peer update ", err.Error())
  89. return
  90. }
  91. logger.Log(1, "no need to update peers")
  92. }()
  93. }
  94. // PublishPeerUpdate --- deterines and publishes a peer update to all the peers of a node
  95. func PublishPeerUpdate(newNode *models.Node) error {
  96. if !servercfg.IsMessageQueueBackend() {
  97. return nil
  98. }
  99. networkNodes, err := logic.GetNetworkNodes(newNode.Network)
  100. if err != nil {
  101. logger.Log(1, "err getting Network Nodes", err.Error())
  102. return err
  103. }
  104. for _, node := range networkNodes {
  105. if node.IsServer == "yes" || node.ID == newNode.ID {
  106. log.Println("skipping update on " + node.Name + " : " + node.ID)
  107. continue
  108. }
  109. log.Println("running update on " + node.Name + " : " + node.ID)
  110. peerUpdate, err := logic.GetPeerUpdate(&node)
  111. if err != nil {
  112. logger.Log(1, "error getting peer update for node", node.ID, err.Error())
  113. continue
  114. }
  115. data, err := json.Marshal(&peerUpdate)
  116. if err != nil {
  117. logger.Log(2, "error marshaling peer update for node", node.ID, err.Error())
  118. continue
  119. }
  120. if err = publish(&node, fmt.Sprintf("peers/%s/%s", node.Network, node.ID), data); err != nil {
  121. logger.Log(1, "failed to publish peer update for node", node.ID)
  122. } else {
  123. logger.Log(1, fmt.Sprintf("sent peer update for node %s on network: %s ", node.Name, node.Network))
  124. }
  125. }
  126. return nil
  127. }
  128. // PublishPeerUpdate --- deterines and publishes a peer update to all the peers of a node
  129. func PublishExtPeerUpdate(node *models.Node) error {
  130. var err error
  131. if logic.IsLocalServer(node) {
  132. if err = logic.ServerUpdate(node, false); err != nil {
  133. logger.Log(1, "server node:", node.ID, "failed to update peers with ext clients")
  134. return err
  135. } else {
  136. return nil
  137. }
  138. }
  139. if !servercfg.IsMessageQueueBackend() {
  140. return nil
  141. }
  142. peerUpdate, err := logic.GetPeerUpdate(node)
  143. if err != nil {
  144. return err
  145. }
  146. data, err := json.Marshal(&peerUpdate)
  147. if err != nil {
  148. return err
  149. }
  150. return publish(node, fmt.Sprintf("peers/%s/%s", node.Network, node.ID), data)
  151. }
  152. // GetID -- decodes a message queue topic and returns the embedded node.ID
  153. func GetID(topic string) (string, error) {
  154. parts := strings.Split(topic, "/")
  155. count := len(parts)
  156. if count == 1 {
  157. return "", errors.New("invalid topic")
  158. }
  159. //the last part of the topic will be the node.ID
  160. return parts[count-1], nil
  161. }
  162. // NodeUpdate -- publishes a node update
  163. func NodeUpdate(node *models.Node) error {
  164. if !servercfg.IsMessageQueueBackend() {
  165. return nil
  166. }
  167. logger.Log(3, "publishing node update to "+node.Name)
  168. data, err := json.Marshal(node)
  169. if err != nil {
  170. logger.Log(2, "error marshalling node update ", err.Error())
  171. return err
  172. }
  173. if err = publish(node, fmt.Sprintf("update/%s/%s", node.Network, node.ID), data); err != nil {
  174. logger.Log(2, "error publishing node update to peer ", node.ID, err.Error())
  175. return err
  176. }
  177. return nil
  178. }
  179. // SetupMQTT creates a connection to broker and return client
  180. func SetupMQTT() mqtt.Client {
  181. opts := mqtt.NewClientOptions()
  182. broker := servercfg.GetMessageQueueEndpoint()
  183. opts.AddBroker(broker)
  184. id := ncutils.MakeRandomString(23)
  185. opts.ClientID = id
  186. client := mqtt.NewClient(opts)
  187. tperiod := time.Now().Add(10 * time.Second)
  188. for {
  189. if token := client.Connect(); token.Wait() && token.Error() != nil {
  190. logger.Log(2, "unable to connect to broker, retrying ...")
  191. if time.Now().After(tperiod) {
  192. log.Fatal(0, "could not connect to broker, exiting ...", token.Error())
  193. }
  194. } else {
  195. break
  196. }
  197. time.Sleep(2 * time.Second)
  198. }
  199. logger.Log(2, "connected to message queue", broker)
  200. return client
  201. }
  202. // Keepalive -- periodically pings all nodes to let them know server is still alive and doing well
  203. func Keepalive(ctx context.Context) {
  204. for {
  205. select {
  206. case <-ctx.Done():
  207. return
  208. case <-time.After(time.Second * KEEPALIVE_TIMEOUT):
  209. client := SetupMQTT()
  210. networks, err := logic.GetNetworks()
  211. if err != nil {
  212. logger.Log(1, "error retrieving networks for keepalive", err.Error())
  213. }
  214. for _, network := range networks {
  215. serverNode, errN := logic.GetNetworkServerLeader(network.NetID)
  216. if errN == nil {
  217. serverNode.SetLastCheckIn()
  218. logic.UpdateNode(&serverNode, &serverNode)
  219. if network.DefaultUDPHolePunch == "yes" {
  220. logic.ShouldPublishPeerPorts(&serverNode)
  221. }
  222. err = PublishPeerUpdate(&serverNode)
  223. if err != nil {
  224. logger.Log(1, "error publishing udp port updates for network", network.NetID)
  225. logger.Log(1, errN.Error())
  226. }
  227. } else {
  228. logger.Log(1, "unable to retrieve leader for network ", network.NetID)
  229. logger.Log(1, errN.Error())
  230. continue
  231. }
  232. if serverNode.Address == "" {
  233. logger.Log(1, "leader not defined for network ", network.NetID)
  234. continue
  235. }
  236. publishServerKeepalive(client, &network)
  237. err = serverctl.SyncServerNetwork(network.NetID)
  238. if err != nil {
  239. logger.Log(1, "error syncing server network", err.Error())
  240. }
  241. }
  242. client.Disconnect(MQ_DISCONNECT)
  243. }
  244. }
  245. }
  246. func publishServerKeepalive(client mqtt.Client, network *models.Network) {
  247. nodes, err := logic.GetNetworkNodes(network.NetID)
  248. if err != nil {
  249. return
  250. }
  251. for _, node := range nodes {
  252. if token := client.Publish(fmt.Sprintf("serverkeepalive/%s/%s", network.NetID, node.ID), 0, false, servercfg.GetVersion()); token.Wait() && token.Error() != nil {
  253. logger.Log(1, "error publishing server keepalive for network", network.NetID, token.Error().Error())
  254. } else {
  255. logger.Log(2, "keepalive sent for network/node", network.NetID, node.ID)
  256. }
  257. }
  258. }