handlers.go 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307
  1. package mq
  2. import (
  3. "encoding/json"
  4. "fmt"
  5. "time"
  6. mqtt "github.com/eclipse/paho.mqtt.golang"
  7. "github.com/gravitl/netmaker/database"
  8. "github.com/gravitl/netmaker/logger"
  9. "github.com/gravitl/netmaker/logic"
  10. "github.com/gravitl/netmaker/models"
  11. "github.com/gravitl/netmaker/netclient/ncutils"
  12. "github.com/gravitl/netmaker/servercfg"
  13. )
  14. // DefaultHandler default message queue handler -- NOT USED
  15. func DefaultHandler(client mqtt.Client, msg mqtt.Message) {
  16. logger.Log(0, "MQTT Message: Topic: ", string(msg.Topic()), " Message: ", string(msg.Payload()))
  17. }
  18. // Ping message Handler -- handles ping topic from client nodes
  19. func Ping(client mqtt.Client, msg mqtt.Message) {
  20. go func() {
  21. id, err := getID(msg.Topic())
  22. if err != nil {
  23. logger.Log(0, "error getting node.ID sent on ping topic ")
  24. return
  25. }
  26. node, err := logic.GetNodeByID(id)
  27. if err != nil {
  28. logger.Log(0, "mq-ping error getting node: ", err.Error())
  29. record, err := database.FetchRecord(database.NODES_TABLE_NAME, id)
  30. if err != nil {
  31. logger.Log(0, "error reading database ", err.Error())
  32. return
  33. }
  34. logger.Log(0, "record from database")
  35. logger.Log(0, record)
  36. return
  37. }
  38. decrypted, decryptErr := decryptMsg(&node, msg.Payload())
  39. if decryptErr != nil {
  40. logger.Log(0, "error decrypting when updating node ", node.ID, decryptErr.Error())
  41. return
  42. }
  43. var checkin models.NodeCheckin
  44. if err := json.Unmarshal(decrypted, &checkin); err != nil {
  45. logger.Log(1, "error unmarshaling payload ", err.Error())
  46. return
  47. }
  48. node.SetLastCheckIn()
  49. node.Version = checkin.Version
  50. node.Connected = checkin.Connected
  51. if err := logic.UpdateNode(&node, &node); err != nil {
  52. logger.Log(0, "error updating node", node.Name, node.ID, " on checkin", err.Error())
  53. return
  54. }
  55. logger.Log(3, "ping processed for node", node.Name, 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. func UpdateNode(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. var newNode models.Node
  79. if err := json.Unmarshal(decrypted, &newNode); err != nil {
  80. logger.Log(1, "error unmarshaling payload ", err.Error())
  81. return
  82. }
  83. ifaceDelta := logic.IfaceDelta(&currentNode, &newNode)
  84. if servercfg.Is_EE && ifaceDelta {
  85. if err = logic.EnterpriseResetAllPeersFailovers(currentNode.ID, currentNode.Network); err != nil {
  86. logger.Log(1, "failed to reset failover list during node update", currentNode.Name, currentNode.Network)
  87. }
  88. }
  89. newNode.SetLastCheckIn()
  90. if err := logic.UpdateNode(&currentNode, &newNode); err != nil {
  91. logger.Log(1, "error saving node", err.Error())
  92. return
  93. }
  94. if ifaceDelta { // reduce number of unneeded updates, by only sending on iface changes
  95. if err = PublishPeerUpdate(&currentNode, true); err != nil {
  96. logger.Log(0, "error updating peers when node", currentNode.Name, currentNode.ID, "informed the server of an interface change", err.Error())
  97. }
  98. }
  99. logger.Log(1, "updated node", id, newNode.Name)
  100. }()
  101. }
  102. // UpdateMetrics message Handler -- handles updates from client nodes for metrics
  103. func UpdateMetrics(client mqtt.Client, msg mqtt.Message) {
  104. if servercfg.Is_EE {
  105. go func() {
  106. id, err := getID(msg.Topic())
  107. if err != nil {
  108. logger.Log(1, "error getting node.ID sent on ", msg.Topic(), err.Error())
  109. return
  110. }
  111. currentNode, err := logic.GetNodeByID(id)
  112. if err != nil {
  113. logger.Log(1, "error getting node ", id, err.Error())
  114. return
  115. }
  116. decrypted, decryptErr := decryptMsg(&currentNode, msg.Payload())
  117. if decryptErr != nil {
  118. logger.Log(1, "failed to decrypt message for node ", id, decryptErr.Error())
  119. return
  120. }
  121. var newMetrics models.Metrics
  122. if err := json.Unmarshal(decrypted, &newMetrics); err != nil {
  123. logger.Log(1, "error unmarshaling payload ", err.Error())
  124. return
  125. }
  126. shouldUpdate := updateNodeMetrics(&currentNode, &newMetrics)
  127. if err = logic.UpdateMetrics(id, &newMetrics); err != nil {
  128. logger.Log(1, "faield to update node metrics", id, currentNode.Name, err.Error())
  129. return
  130. }
  131. if servercfg.IsMetricsExporter() {
  132. if err := pushMetricsToExporter(newMetrics); err != nil {
  133. logger.Log(2, fmt.Sprintf("failed to push node: [%s] metrics to exporter, err: %v",
  134. currentNode.Name, err))
  135. }
  136. }
  137. if newMetrics.Connectivity != nil {
  138. err := logic.EnterpriseFailoverFunc(&currentNode)
  139. if err != nil {
  140. logger.Log(0, "failed to failover for node", currentNode.Name, "on network", currentNode.Network, "-", err.Error())
  141. }
  142. }
  143. if shouldUpdate {
  144. logger.Log(2, "updating peers after node", currentNode.Name, currentNode.Network, "detected connectivity issues")
  145. if err = PublishSinglePeerUpdate(&currentNode); err != nil {
  146. logger.Log(0, "failed to publish update after failover peer change for node", currentNode.Name, currentNode.Network)
  147. }
  148. }
  149. logger.Log(1, "updated node metrics", id, currentNode.Name)
  150. }()
  151. }
  152. }
  153. // ClientPeerUpdate message handler -- handles updating peers after signal from client nodes
  154. func ClientPeerUpdate(client mqtt.Client, msg mqtt.Message) {
  155. go func() {
  156. id, err := getID(msg.Topic())
  157. if err != nil {
  158. logger.Log(1, "error getting node.ID sent on ", msg.Topic(), err.Error())
  159. return
  160. }
  161. currentNode, err := logic.GetNodeByID(id)
  162. if err != nil {
  163. logger.Log(1, "error getting node ", id, err.Error())
  164. return
  165. }
  166. decrypted, decryptErr := decryptMsg(&currentNode, msg.Payload())
  167. if decryptErr != nil {
  168. logger.Log(1, "failed to decrypt message during client peer update for node ", id, decryptErr.Error())
  169. return
  170. }
  171. switch decrypted[0] {
  172. case ncutils.ACK:
  173. currentServerNode, err := logic.GetNetworkServerLocal(currentNode.Network)
  174. if err != nil {
  175. return
  176. }
  177. if err := logic.ServerUpdate(&currentServerNode, false); err != nil {
  178. logger.Log(1, "server node:", currentServerNode.ID, "failed update")
  179. return
  180. }
  181. case ncutils.DONE:
  182. updateNodePeers(&currentNode)
  183. }
  184. logger.Log(1, "sent peer updates after signal received from", id, currentNode.Name)
  185. }()
  186. }
  187. func updateNodePeers(currentNode *models.Node) {
  188. currentServerNode, err := logic.GetNetworkServerLocal(currentNode.Network)
  189. if err != nil {
  190. logger.Log(1, "failed to get server node failed update\n", err.Error())
  191. return
  192. }
  193. if err := logic.ServerUpdate(&currentServerNode, false); err != nil {
  194. logger.Log(1, "server node:", currentServerNode.ID, "failed update")
  195. return
  196. }
  197. if logic.IsLeader(&currentServerNode) {
  198. if err := PublishPeerUpdate(currentNode, false); err != nil {
  199. logger.Log(1, "error publishing peer update ", err.Error())
  200. return
  201. }
  202. }
  203. }
  204. func updateNodeMetrics(currentNode *models.Node, newMetrics *models.Metrics) bool {
  205. if newMetrics.FailoverPeers == nil {
  206. newMetrics.FailoverPeers = make(map[string]string)
  207. }
  208. oldMetrics, err := logic.GetMetrics(currentNode.ID)
  209. if err != nil {
  210. logger.Log(1, "error finding old metrics for node", currentNode.ID, currentNode.Name)
  211. return false
  212. }
  213. if oldMetrics.FailoverPeers == nil {
  214. oldMetrics.FailoverPeers = make(map[string]string)
  215. }
  216. var attachedClients []models.ExtClient
  217. if currentNode.IsIngressGateway == "yes" {
  218. clients, err := logic.GetExtClientsByID(currentNode.ID, currentNode.Network)
  219. if err == nil {
  220. attachedClients = clients
  221. }
  222. }
  223. if len(attachedClients) > 0 {
  224. // associate ext clients with IDs
  225. for i := range attachedClients {
  226. extMetric := newMetrics.Connectivity[attachedClients[i].PublicKey]
  227. if len(extMetric.NodeName) == 0 &&
  228. len(newMetrics.Connectivity[attachedClients[i].ClientID].NodeName) > 0 { // cover server clients
  229. extMetric = newMetrics.Connectivity[attachedClients[i].ClientID]
  230. if extMetric.TotalReceived > 0 && extMetric.TotalSent > 0 {
  231. extMetric.Connected = true
  232. }
  233. }
  234. extMetric.NodeName = attachedClients[i].ClientID
  235. extMetric.IsServer = "no"
  236. delete(newMetrics.Connectivity, attachedClients[i].PublicKey)
  237. newMetrics.Connectivity[attachedClients[i].ClientID] = extMetric
  238. }
  239. }
  240. // run through metrics for each peer
  241. for k := range newMetrics.Connectivity {
  242. currMetric := newMetrics.Connectivity[k]
  243. oldMetric := oldMetrics.Connectivity[k]
  244. currMetric.TotalTime += oldMetric.TotalTime
  245. currMetric.Uptime += oldMetric.Uptime // get the total uptime for this connection
  246. if currMetric.Uptime == 0 || currMetric.TotalTime == 0 {
  247. currMetric.PercentUp = 0
  248. } else {
  249. currMetric.PercentUp = 100.0 * (float64(currMetric.Uptime) / float64(currMetric.TotalTime))
  250. }
  251. totalUpMinutes := currMetric.Uptime * ncutils.CheckInInterval
  252. currMetric.ActualUptime = time.Duration(totalUpMinutes) * time.Minute
  253. delete(oldMetrics.Connectivity, k) // remove from old data
  254. newMetrics.Connectivity[k] = currMetric
  255. }
  256. // add nodes that need failover
  257. nodes, err := logic.GetNetworkNodes(currentNode.Network)
  258. if err != nil {
  259. logger.Log(0, "failed to retrieve nodes while updating metrics")
  260. return false
  261. }
  262. for _, node := range nodes {
  263. if !newMetrics.Connectivity[node.ID].Connected &&
  264. len(newMetrics.Connectivity[node.ID].NodeName) > 0 &&
  265. node.Connected == "yes" &&
  266. len(node.FailoverNode) > 0 &&
  267. node.Failover != "yes" {
  268. newMetrics.FailoverPeers[node.ID] = node.FailoverNode
  269. }
  270. }
  271. shouldUpdate := len(oldMetrics.FailoverPeers) == 0 && len(newMetrics.FailoverPeers) > 0
  272. for k, v := range oldMetrics.FailoverPeers {
  273. if len(newMetrics.FailoverPeers[k]) > 0 && len(v) == 0 {
  274. shouldUpdate = true
  275. }
  276. if len(v) > 0 && len(newMetrics.FailoverPeers[k]) == 0 {
  277. newMetrics.FailoverPeers[k] = v
  278. }
  279. }
  280. for k := range oldMetrics.Connectivity { // cleanup any left over data, self healing
  281. delete(newMetrics.Connectivity, k)
  282. }
  283. return shouldUpdate
  284. }