handlers.go 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306
  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.String(), 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. host, err := logic.GetHost(node.HostID.String())
  49. if err != nil {
  50. logger.Log(0, "error retrieving host for node ", node.ID.String(), err.Error())
  51. return
  52. }
  53. node.SetLastCheckIn()
  54. host.Version = checkin.Version
  55. node.Connected = checkin.Connected
  56. host.Interfaces = checkin.Ifaces
  57. for i := range host.Interfaces {
  58. host.Interfaces[i].AddressString = host.Interfaces[i].Address.String()
  59. }
  60. if err := logic.UpdateNode(&node, &node); err != nil {
  61. logger.Log(0, "error updating node", node.ID.String(), " on checkin", err.Error())
  62. return
  63. }
  64. logger.Log(3, "ping processed for node", node.ID.String())
  65. // --TODO --set client version once feature is implemented.
  66. //node.SetClientVersion(msg.Payload())
  67. }()
  68. }
  69. // UpdateNode message Handler -- handles updates from client nodes
  70. func UpdateNode(client mqtt.Client, msg mqtt.Message) {
  71. go func() {
  72. id, err := getID(msg.Topic())
  73. if err != nil {
  74. logger.Log(1, "error getting node.ID sent on ", msg.Topic(), err.Error())
  75. return
  76. }
  77. currentNode, err := logic.GetNodeByID(id)
  78. if err != nil {
  79. logger.Log(1, "error getting node ", id, err.Error())
  80. return
  81. }
  82. decrypted, decryptErr := decryptMsg(&currentNode, msg.Payload())
  83. if decryptErr != nil {
  84. logger.Log(1, "failed to decrypt message for node ", id, decryptErr.Error())
  85. return
  86. }
  87. var newNode models.Node
  88. if err := json.Unmarshal(decrypted, &newNode); err != nil {
  89. logger.Log(1, "error unmarshaling payload ", err.Error())
  90. return
  91. }
  92. ifaceDelta := logic.IfaceDelta(&currentNode, &newNode)
  93. if servercfg.Is_EE && ifaceDelta {
  94. if err = logic.EnterpriseResetAllPeersFailovers(currentNode.ID.String(), currentNode.Network); err != nil {
  95. logger.Log(1, "failed to reset failover list during node update", currentNode.ID.String(), currentNode.Network)
  96. }
  97. }
  98. newNode.SetLastCheckIn()
  99. if err := logic.UpdateNode(&currentNode, &newNode); err != nil {
  100. logger.Log(1, "error saving node", err.Error())
  101. return
  102. }
  103. if ifaceDelta { // reduce number of unneeded updates, by only sending on iface changes
  104. if err = PublishPeerUpdate(currentNode.Network, true); err != nil {
  105. logger.Log(0, "error updating peers when node", currentNode.ID.String(), "informed the server of an interface change", err.Error())
  106. }
  107. }
  108. logger.Log(1, "updated node", id, newNode.ID.String())
  109. }()
  110. }
  111. // UpdateMetrics message Handler -- handles updates from client nodes for metrics
  112. func UpdateMetrics(client mqtt.Client, msg mqtt.Message) {
  113. if servercfg.Is_EE {
  114. go func() {
  115. id, err := getID(msg.Topic())
  116. if err != nil {
  117. logger.Log(1, "error getting node.ID sent on ", msg.Topic(), err.Error())
  118. return
  119. }
  120. currentNode, err := logic.GetNodeByID(id)
  121. if err != nil {
  122. logger.Log(1, "error getting node ", id, err.Error())
  123. return
  124. }
  125. decrypted, decryptErr := decryptMsg(&currentNode, msg.Payload())
  126. if decryptErr != nil {
  127. logger.Log(1, "failed to decrypt message for node ", id, decryptErr.Error())
  128. return
  129. }
  130. var newMetrics models.Metrics
  131. if err := json.Unmarshal(decrypted, &newMetrics); err != nil {
  132. logger.Log(1, "error unmarshaling payload ", err.Error())
  133. return
  134. }
  135. shouldUpdate := updateNodeMetrics(&currentNode, &newMetrics)
  136. if err = logic.UpdateMetrics(id, &newMetrics); err != nil {
  137. logger.Log(1, "faield to update node metrics", id, err.Error())
  138. return
  139. }
  140. if servercfg.IsMetricsExporter() {
  141. if err := pushMetricsToExporter(newMetrics); err != nil {
  142. logger.Log(2, fmt.Sprintf("failed to push node: [%s] metrics to exporter, err: %v",
  143. currentNode.ID, err))
  144. }
  145. }
  146. if newMetrics.Connectivity != nil {
  147. err := logic.EnterpriseFailoverFunc(&currentNode)
  148. if err != nil {
  149. logger.Log(0, "failed to failover for node", currentNode.ID.String(), "on network", currentNode.Network, "-", err.Error())
  150. }
  151. }
  152. if shouldUpdate {
  153. logger.Log(2, "updating peers after node", currentNode.ID.String(), currentNode.Network, "detected connectivity issues")
  154. if err = PublishSinglePeerUpdate(&currentNode); err != nil {
  155. logger.Log(0, "failed to publish update after failover peer change for node", currentNode.ID.String(), currentNode.Network)
  156. }
  157. }
  158. logger.Log(1, "updated node metrics", id)
  159. }()
  160. }
  161. }
  162. // ClientPeerUpdate message handler -- handles updating peers after signal from client nodes
  163. func ClientPeerUpdate(client mqtt.Client, msg mqtt.Message) {
  164. go func() {
  165. id, err := getID(msg.Topic())
  166. if err != nil {
  167. logger.Log(1, "error getting node.ID sent on ", msg.Topic(), err.Error())
  168. return
  169. }
  170. currentNode, err := logic.GetNodeByID(id)
  171. if err != nil {
  172. logger.Log(1, "error getting node ", id, err.Error())
  173. return
  174. }
  175. decrypted, decryptErr := decryptMsg(&currentNode, msg.Payload())
  176. if decryptErr != nil {
  177. logger.Log(1, "failed to decrypt message during client peer update for node ", id, decryptErr.Error())
  178. return
  179. }
  180. switch decrypted[0] {
  181. case ncutils.ACK:
  182. //do we still need this
  183. case ncutils.DONE:
  184. updateNodePeers(&currentNode)
  185. }
  186. logger.Log(1, "sent peer updates after signal received from", id)
  187. }()
  188. }
  189. func updateNodePeers(currentNode *models.Node) {
  190. if err := PublishPeerUpdate(currentNode.Network, false); err != nil {
  191. logger.Log(1, "error publishing peer update ", err.Error())
  192. return
  193. }
  194. }
  195. func updateNodeMetrics(currentNode *models.Node, newMetrics *models.Metrics) bool {
  196. if newMetrics.FailoverPeers == nil {
  197. newMetrics.FailoverPeers = make(map[string]string)
  198. }
  199. oldMetrics, err := logic.GetMetrics(currentNode.ID.String())
  200. if err != nil {
  201. logger.Log(1, "error finding old metrics for node", currentNode.ID.String())
  202. return false
  203. }
  204. if oldMetrics.FailoverPeers == nil {
  205. oldMetrics.FailoverPeers = make(map[string]string)
  206. }
  207. var attachedClients []models.ExtClient
  208. if currentNode.IsIngressGateway {
  209. clients, err := logic.GetExtClientsByID(currentNode.ID.String(), currentNode.Network)
  210. if err == nil {
  211. attachedClients = clients
  212. }
  213. }
  214. if len(attachedClients) > 0 {
  215. // associate ext clients with IDs
  216. for i := range attachedClients {
  217. extMetric := newMetrics.Connectivity[attachedClients[i].PublicKey]
  218. if len(extMetric.NodeName) == 0 &&
  219. len(newMetrics.Connectivity[attachedClients[i].ClientID].NodeName) > 0 { // cover server clients
  220. extMetric = newMetrics.Connectivity[attachedClients[i].ClientID]
  221. if extMetric.TotalReceived > 0 && extMetric.TotalSent > 0 {
  222. extMetric.Connected = true
  223. }
  224. }
  225. extMetric.NodeName = attachedClients[i].ClientID
  226. extMetric.IsServer = "no"
  227. delete(newMetrics.Connectivity, attachedClients[i].PublicKey)
  228. newMetrics.Connectivity[attachedClients[i].ClientID] = extMetric
  229. }
  230. }
  231. // run through metrics for each peer
  232. for k := range newMetrics.Connectivity {
  233. currMetric := newMetrics.Connectivity[k]
  234. oldMetric := oldMetrics.Connectivity[k]
  235. currMetric.TotalTime += oldMetric.TotalTime
  236. currMetric.Uptime += oldMetric.Uptime // get the total uptime for this connection
  237. if currMetric.Uptime == 0 || currMetric.TotalTime == 0 {
  238. currMetric.PercentUp = 0
  239. } else {
  240. currMetric.PercentUp = 100.0 * (float64(currMetric.Uptime) / float64(currMetric.TotalTime))
  241. }
  242. totalUpMinutes := currMetric.Uptime * ncutils.CheckInInterval
  243. currMetric.ActualUptime = time.Duration(totalUpMinutes) * time.Minute
  244. delete(oldMetrics.Connectivity, k) // remove from old data
  245. newMetrics.Connectivity[k] = currMetric
  246. if oldProxyMetric, ok := oldMetrics.ProxyMetrics[k]; ok {
  247. newProxyMetric := newMetrics.ProxyMetrics[k]
  248. newProxyMetric.TrafficSent += oldProxyMetric.TrafficSent
  249. newProxyMetric.TrafficRecieved += oldProxyMetric.TrafficRecieved
  250. newMetrics.ProxyMetrics[k] = newProxyMetric
  251. }
  252. }
  253. // add nodes that need failover
  254. nodes, err := logic.GetNetworkNodes(currentNode.Network)
  255. if err != nil {
  256. logger.Log(0, "failed to retrieve nodes while updating metrics")
  257. return false
  258. }
  259. for _, node := range nodes {
  260. if !newMetrics.Connectivity[node.ID.String()].Connected &&
  261. len(newMetrics.Connectivity[node.ID.String()].NodeName) > 0 &&
  262. node.Connected == true &&
  263. len(node.FailoverNode) > 0 &&
  264. !node.Failover {
  265. newMetrics.FailoverPeers[node.ID.String()] = node.FailoverNode.String()
  266. }
  267. }
  268. shouldUpdate := len(oldMetrics.FailoverPeers) == 0 && len(newMetrics.FailoverPeers) > 0
  269. for k, v := range oldMetrics.FailoverPeers {
  270. if len(newMetrics.FailoverPeers[k]) > 0 && len(v) == 0 {
  271. shouldUpdate = true
  272. }
  273. if len(v) > 0 && len(newMetrics.FailoverPeers[k]) == 0 {
  274. newMetrics.FailoverPeers[k] = v
  275. }
  276. }
  277. for k := range oldMetrics.Connectivity { // cleanup any left over data, self healing
  278. delete(newMetrics.Connectivity, k)
  279. }
  280. return shouldUpdate
  281. }