mqpublish.go 9.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304
  1. package functions
  2. import (
  3. "context"
  4. "encoding/json"
  5. "errors"
  6. "fmt"
  7. "io"
  8. "net"
  9. "net/http"
  10. "strconv"
  11. "sync"
  12. "time"
  13. "github.com/cloverstd/tcping/ping"
  14. "github.com/gravitl/netmaker/logger"
  15. "github.com/gravitl/netmaker/logic/metrics"
  16. "github.com/gravitl/netmaker/models"
  17. "github.com/gravitl/netmaker/netclient/auth"
  18. "github.com/gravitl/netmaker/netclient/config"
  19. "github.com/gravitl/netmaker/netclient/ncutils"
  20. )
  21. var metricsCache = new(sync.Map)
  22. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  23. //
  24. // if there are no updates, simply "pings" the server as a checkin
  25. func Checkin(ctx context.Context, wg *sync.WaitGroup) {
  26. logger.Log(2, "starting checkin goroutine")
  27. defer wg.Done()
  28. ticker := time.NewTicker(time.Minute * ncutils.CheckInInterval)
  29. defer ticker.Stop()
  30. for {
  31. select {
  32. case <-ctx.Done():
  33. logger.Log(0, "checkin routine closed")
  34. return
  35. case <-ticker.C:
  36. if mqclient != nil && mqclient.IsConnected() {
  37. checkin()
  38. } else {
  39. logger.Log(0, "MQ client is not connected, skipping checkin...")
  40. }
  41. }
  42. }
  43. }
  44. func checkin() {
  45. networks, _ := ncutils.GetSystemNetworks()
  46. logger.Log(3, "checkin with server(s) for all networks")
  47. for _, network := range networks {
  48. var nodeCfg config.ClientConfig
  49. nodeCfg.Network = network
  50. nodeCfg.ReadConfig()
  51. // check for nftables present if on Linux
  52. if ncutils.IsLinux() {
  53. if ncutils.IsNFTablesPresent() {
  54. nodeCfg.Node.FirewallInUse = models.FIREWALL_NFTABLES
  55. } else {
  56. nodeCfg.Node.FirewallInUse = models.FIREWALL_IPTABLES
  57. }
  58. } else {
  59. // defaults to iptables for now, may need another default for non-Linux OSes
  60. nodeCfg.Node.FirewallInUse = models.FIREWALL_IPTABLES
  61. }
  62. if nodeCfg.Node.Connected == "yes" {
  63. if nodeCfg.Node.IsStatic != "yes" {
  64. extIP, err := ncutils.GetPublicIP(nodeCfg.Server.API)
  65. if err != nil {
  66. logger.Log(1, "error encountered checking public ip addresses: ", err.Error())
  67. }
  68. if nodeCfg.Node.Endpoint != extIP && extIP != "" {
  69. logger.Log(1, "network:", nodeCfg.Node.Network, "endpoint has changed from ", nodeCfg.Node.Endpoint, " to ", extIP)
  70. nodeCfg.Node.Endpoint = extIP
  71. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  72. logger.Log(0, "network:", nodeCfg.Node.Network, "could not publish endpoint change")
  73. }
  74. }
  75. intIP, err := getPrivateAddr()
  76. if err != nil {
  77. logger.Log(1, "network:", nodeCfg.Node.Network, "error encountered checking private ip addresses: ", err.Error())
  78. }
  79. if nodeCfg.Node.LocalAddress != intIP && intIP != "" {
  80. logger.Log(1, "network:", nodeCfg.Node.Network, "local Address has changed from ", nodeCfg.Node.LocalAddress, " to ", intIP)
  81. nodeCfg.Node.LocalAddress = intIP
  82. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  83. logger.Log(0, "Network: ", nodeCfg.Node.Network, " could not publish local address change")
  84. }
  85. }
  86. _ = UpdateLocalListenPort(&nodeCfg)
  87. } else if nodeCfg.Node.IsLocal == "yes" && nodeCfg.Node.LocalRange != "" {
  88. localIP, err := ncutils.GetLocalIP(nodeCfg.Node.LocalRange)
  89. if err != nil {
  90. logger.Log(1, "network:", nodeCfg.Node.Network, "error encountered checking local ip addresses: ", err.Error())
  91. }
  92. if nodeCfg.Node.Endpoint != localIP && localIP != "" {
  93. logger.Log(1, "network:", nodeCfg.Node.Network, "endpoint has changed from "+nodeCfg.Node.Endpoint+" to ", localIP)
  94. nodeCfg.Node.Endpoint = localIP
  95. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  96. logger.Log(0, "network:", nodeCfg.Node.Network, "could not publish localip change")
  97. }
  98. }
  99. }
  100. }
  101. //check version
  102. if nodeCfg.Node.Version != ncutils.Version {
  103. nodeCfg.Node.Version = ncutils.Version
  104. config.Write(&nodeCfg, nodeCfg.Network)
  105. }
  106. Hello(&nodeCfg)
  107. if nodeCfg.Server.Is_EE && nodeCfg.Node.Connected == "yes" {
  108. logger.Log(0, "collecting metrics for node", nodeCfg.Node.Name)
  109. publishMetrics(&nodeCfg)
  110. }
  111. }
  112. }
  113. // PublishNodeUpdates -- saves node and pushes changes to broker
  114. func PublishNodeUpdate(nodeCfg *config.ClientConfig) error {
  115. if err := config.Write(nodeCfg, nodeCfg.Network); err != nil {
  116. return err
  117. }
  118. data, err := json.Marshal(nodeCfg.Node)
  119. if err != nil {
  120. return err
  121. }
  122. if err = publish(nodeCfg, fmt.Sprintf("update/%s", nodeCfg.Node.ID), data, 1); err != nil {
  123. return err
  124. }
  125. logger.Log(0, "network:", nodeCfg.Node.Network, "sent a node update to server for node", nodeCfg.Node.Name, ", ", nodeCfg.Node.ID)
  126. return nil
  127. }
  128. // Hello -- ping the broker to let server know node it's alive and well
  129. func Hello(nodeCfg *config.ClientConfig) {
  130. var checkin models.NodeCheckin
  131. checkin.Version = ncutils.Version
  132. checkin.Connected = nodeCfg.Node.Connected
  133. ip, err := getInterfaces()
  134. if err != nil {
  135. logger.Log(0, "failed to retrieve local interfaces", err.Error())
  136. } else {
  137. nodeCfg.Node.Interfaces = *ip
  138. config.Write(nodeCfg, nodeCfg.Network)
  139. }
  140. checkin.Ifaces = nodeCfg.Node.Interfaces
  141. data, err := json.Marshal(checkin)
  142. if err != nil {
  143. logger.Log(0, "unable to marshal checkin data", err.Error())
  144. return
  145. }
  146. if err := publish(nodeCfg, fmt.Sprintf("ping/%s", nodeCfg.Node.ID), data, 0); err != nil {
  147. logger.Log(0, fmt.Sprintf("Network: %s error publishing ping, %v", nodeCfg.Node.Network, err))
  148. logger.Log(0, "running pull on "+nodeCfg.Node.Network+" to reconnect")
  149. _, err := Pull(nodeCfg.Node.Network, true)
  150. if err != nil {
  151. logger.Log(0, "could not run pull on "+nodeCfg.Node.Network+", error: "+err.Error())
  152. }
  153. } else {
  154. logger.Log(3, "checkin for", nodeCfg.Network, "complete")
  155. }
  156. }
  157. // publishMetrics - publishes the metrics of a given nodecfg
  158. func publishMetrics(nodeCfg *config.ClientConfig) {
  159. token, err := Authenticate(nodeCfg)
  160. if err != nil {
  161. logger.Log(1, "failed to authenticate when publishing metrics", err.Error())
  162. return
  163. }
  164. url := fmt.Sprintf("https://%s/api/nodes/%s/%s", nodeCfg.Server.API, nodeCfg.Network, nodeCfg.Node.ID)
  165. response, err := API("", http.MethodGet, url, token)
  166. if err != nil {
  167. logger.Log(1, "failed to read from server during metrics publish", err.Error())
  168. return
  169. }
  170. if response.StatusCode != http.StatusOK {
  171. bytes, err := io.ReadAll(response.Body)
  172. if err != nil {
  173. fmt.Println(err)
  174. }
  175. logger.Log(0, fmt.Sprintf("%s %s", string(bytes), err.Error()))
  176. return
  177. }
  178. defer response.Body.Close()
  179. var nodeGET models.NodeGet
  180. if err := json.NewDecoder(response.Body).Decode(&nodeGET); err != nil {
  181. logger.Log(0, "failed to decode node when running metrics update", err.Error())
  182. return
  183. }
  184. metrics, err := metrics.Collect(nodeCfg.Node.Interface, nodeGET.PeerIDs)
  185. if err != nil {
  186. logger.Log(0, "failed metric collection for node", nodeCfg.Node.Name, err.Error())
  187. }
  188. metrics.Network = nodeCfg.Node.Network
  189. metrics.NodeName = nodeCfg.Node.Name
  190. metrics.NodeID = nodeCfg.Node.ID
  191. metrics.IsServer = "no"
  192. data, err := json.Marshal(metrics)
  193. if err != nil {
  194. logger.Log(0, "something went wrong when marshalling metrics data for node", nodeCfg.Node.Name, err.Error())
  195. }
  196. if err = publish(nodeCfg, fmt.Sprintf("metrics/%s", nodeCfg.Node.ID), data, 1); err != nil {
  197. logger.Log(0, "error occurred during publishing of metrics on node", nodeCfg.Node.Name, err.Error())
  198. logger.Log(0, "aggregating metrics locally until broker connection re-established")
  199. val, ok := metricsCache.Load(nodeCfg.Node.ID)
  200. if !ok {
  201. metricsCache.Store(nodeCfg.Node.ID, data)
  202. } else {
  203. var oldMetrics models.Metrics
  204. err = json.Unmarshal(val.([]byte), &oldMetrics)
  205. if err == nil {
  206. for k := range oldMetrics.Connectivity {
  207. currentMetric := metrics.Connectivity[k]
  208. if currentMetric.Latency == 0 {
  209. currentMetric.Latency = oldMetrics.Connectivity[k].Latency
  210. }
  211. currentMetric.Uptime += oldMetrics.Connectivity[k].Uptime
  212. currentMetric.TotalTime += oldMetrics.Connectivity[k].TotalTime
  213. metrics.Connectivity[k] = currentMetric
  214. }
  215. newData, err := json.Marshal(metrics)
  216. if err == nil {
  217. metricsCache.Store(nodeCfg.Node.ID, newData)
  218. }
  219. }
  220. }
  221. } else {
  222. metricsCache.Delete(nodeCfg.Node.ID)
  223. logger.Log(0, "published metrics for node", nodeCfg.Node.Name)
  224. }
  225. }
  226. // node cfg is required in order to fetch the traffic keys of that node for encryption
  227. func publish(nodeCfg *config.ClientConfig, dest string, msg []byte, qos byte) error {
  228. // setup the keys
  229. trafficPrivKey, err := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  230. if err != nil {
  231. return err
  232. }
  233. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  234. if err != nil {
  235. return err
  236. }
  237. encrypted, err := ncutils.Chunk(msg, serverPubKey, trafficPrivKey)
  238. if err != nil {
  239. return err
  240. }
  241. if mqclient == nil {
  242. return errors.New("unable to publish ... no mqclient")
  243. }
  244. if token := mqclient.Publish(dest, qos, false, encrypted); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
  245. logger.Log(0, "could not connect to broker at "+nodeCfg.Server.Server+":"+nodeCfg.Server.MQPort)
  246. var err error
  247. if token.Error() == nil {
  248. err = errors.New("connection timeout")
  249. } else {
  250. err = token.Error()
  251. }
  252. if err != nil {
  253. return err
  254. }
  255. }
  256. return nil
  257. }
  258. func checkBroker(broker string, port string) error {
  259. if broker == "" {
  260. return errors.New("error: broker address is blank")
  261. }
  262. if port == "" {
  263. return errors.New("error: broker port is blank")
  264. }
  265. _, err := net.LookupIP(broker)
  266. if err != nil {
  267. return errors.New("nslookup failed for broker ... check dns records")
  268. }
  269. pinger := ping.NewTCPing()
  270. intPort, err := strconv.Atoi(port)
  271. if err != nil {
  272. logger.Log(1, "error converting port to int: "+err.Error())
  273. }
  274. pinger.SetTarget(&ping.Target{
  275. Protocol: ping.TCP,
  276. Host: broker,
  277. Port: intPort,
  278. Counter: 3,
  279. Interval: 1 * time.Second,
  280. Timeout: 2 * time.Second,
  281. })
  282. pingerDone := pinger.Start()
  283. <-pingerDone
  284. if pinger.Result().SuccessCounter == 0 {
  285. return errors.New("unable to connect to broker port ... check netmaker server and firewalls")
  286. }
  287. return nil
  288. }