mqpublish.go 9.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292
  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. checkin()
  29. ticker := time.NewTicker(time.Minute * ncutils.CheckInInterval)
  30. defer ticker.Stop()
  31. for {
  32. select {
  33. case <-ctx.Done():
  34. logger.Log(0, "checkin routine closed")
  35. return
  36. case <-ticker.C:
  37. checkin()
  38. }
  39. }
  40. }
  41. func checkin() {
  42. networks, _ := ncutils.GetSystemNetworks()
  43. logger.Log(3, "checkin with server(s) for all networks")
  44. for _, network := range networks {
  45. var nodeCfg config.ClientConfig
  46. nodeCfg.Network = network
  47. nodeCfg.ReadConfig()
  48. // check for nftables present if on Linux
  49. if ncutils.IsLinux() {
  50. if ncutils.IsNFTablesPresent() {
  51. nodeCfg.Node.FirewallInUse = models.FIREWALL_NFTABLES
  52. } else {
  53. nodeCfg.Node.FirewallInUse = models.FIREWALL_IPTABLES
  54. }
  55. } else {
  56. // defaults to iptables for now, may need another default for non-Linux OSes
  57. nodeCfg.Node.FirewallInUse = models.FIREWALL_IPTABLES
  58. }
  59. if nodeCfg.Node.Connected == "yes" {
  60. if nodeCfg.Node.IsStatic != "yes" {
  61. extIP, err := ncutils.GetPublicIP(nodeCfg.Server.API)
  62. if err != nil {
  63. logger.Log(1, "error encountered checking public ip addresses: ", err.Error())
  64. }
  65. if nodeCfg.Node.Endpoint != extIP && extIP != "" {
  66. logger.Log(1, "network:", nodeCfg.Node.Network, "endpoint has changed from ", nodeCfg.Node.Endpoint, " to ", extIP)
  67. nodeCfg.Node.Endpoint = extIP
  68. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  69. logger.Log(0, "network:", nodeCfg.Node.Network, "could not publish endpoint change")
  70. }
  71. }
  72. intIP, err := getPrivateAddr()
  73. if err != nil {
  74. logger.Log(1, "network:", nodeCfg.Node.Network, "error encountered checking private ip addresses: ", err.Error())
  75. }
  76. if nodeCfg.Node.LocalAddress != intIP && intIP != "" {
  77. logger.Log(1, "network:", nodeCfg.Node.Network, "local Address has changed from ", nodeCfg.Node.LocalAddress, " to ", intIP)
  78. nodeCfg.Node.LocalAddress = intIP
  79. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  80. logger.Log(0, "Network: ", nodeCfg.Node.Network, " could not publish local address change")
  81. }
  82. }
  83. _ = UpdateLocalListenPort(&nodeCfg)
  84. } else if nodeCfg.Node.IsLocal == "yes" && nodeCfg.Node.LocalRange != "" {
  85. localIP, err := ncutils.GetLocalIP(nodeCfg.Node.LocalRange)
  86. if err != nil {
  87. logger.Log(1, "network:", nodeCfg.Node.Network, "error encountered checking local ip addresses: ", err.Error())
  88. }
  89. if nodeCfg.Node.Endpoint != localIP && localIP != "" {
  90. logger.Log(1, "network:", nodeCfg.Node.Network, "endpoint has changed from "+nodeCfg.Node.Endpoint+" to ", localIP)
  91. nodeCfg.Node.Endpoint = localIP
  92. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  93. logger.Log(0, "network:", nodeCfg.Node.Network, "could not publish localip change")
  94. }
  95. }
  96. }
  97. }
  98. //check version
  99. if nodeCfg.Node.Version != ncutils.Version {
  100. nodeCfg.Node.Version = ncutils.Version
  101. config.Write(&nodeCfg, nodeCfg.Network)
  102. }
  103. Hello(&nodeCfg)
  104. if nodeCfg.Server.Is_EE {
  105. logger.Log(0, "collecting metrics for node", nodeCfg.Node.Name)
  106. publishMetrics(&nodeCfg)
  107. }
  108. }
  109. }
  110. // PublishNodeUpdates -- saves node and pushes changes to broker
  111. func PublishNodeUpdate(nodeCfg *config.ClientConfig) error {
  112. if err := config.Write(nodeCfg, nodeCfg.Network); err != nil {
  113. return err
  114. }
  115. data, err := json.Marshal(nodeCfg.Node)
  116. if err != nil {
  117. return err
  118. }
  119. if err = publish(nodeCfg, fmt.Sprintf("update/%s", nodeCfg.Node.ID), data, 1); err != nil {
  120. return err
  121. }
  122. logger.Log(0, "network:", nodeCfg.Node.Network, "sent a node update to server for node", nodeCfg.Node.Name, ", ", nodeCfg.Node.ID)
  123. return nil
  124. }
  125. // Hello -- ping the broker to let server know node it's alive and well
  126. func Hello(nodeCfg *config.ClientConfig) {
  127. var checkin models.NodeCheckin
  128. checkin.Version = ncutils.Version
  129. checkin.Connected = nodeCfg.Node.Connected
  130. data, err := json.Marshal(checkin)
  131. if err != nil {
  132. logger.Log(0, "unable to marshal checkin data", err.Error())
  133. return
  134. }
  135. if err := publish(nodeCfg, fmt.Sprintf("ping/%s", nodeCfg.Node.ID), data, 0); err != nil {
  136. logger.Log(0, fmt.Sprintf("Network: %s error publishing ping, %v", nodeCfg.Node.Network, err))
  137. logger.Log(0, "running pull on "+nodeCfg.Node.Network+" to reconnect")
  138. _, err := Pull(nodeCfg.Node.Network, true)
  139. if err != nil {
  140. logger.Log(0, "could not run pull on "+nodeCfg.Node.Network+", error: "+err.Error())
  141. }
  142. } else {
  143. logger.Log(3, "checkin for", nodeCfg.Network, "complete")
  144. }
  145. }
  146. // publishMetrics - publishes the metrics of a given nodecfg
  147. func publishMetrics(nodeCfg *config.ClientConfig) {
  148. token, err := Authenticate(nodeCfg)
  149. if err != nil {
  150. logger.Log(1, "failed to authenticate when publishing metrics", err.Error())
  151. return
  152. }
  153. url := fmt.Sprintf("https://%s/api/nodes/%s/%s", nodeCfg.Server.API, nodeCfg.Network, nodeCfg.Node.ID)
  154. response, err := API("", http.MethodGet, url, token)
  155. if err != nil {
  156. logger.Log(1, "failed to read from server during metrics publish", err.Error())
  157. return
  158. }
  159. if response.StatusCode != http.StatusOK {
  160. bytes, err := io.ReadAll(response.Body)
  161. if err != nil {
  162. fmt.Println(err)
  163. }
  164. logger.Log(0, fmt.Sprintf("%s %s", string(bytes), err.Error()))
  165. return
  166. }
  167. defer response.Body.Close()
  168. var nodeGET models.NodeGet
  169. if err := json.NewDecoder(response.Body).Decode(&nodeGET); err != nil {
  170. logger.Log(0, "failed to decode node when running metrics update", err.Error())
  171. return
  172. }
  173. metrics, err := metrics.Collect(nodeCfg.Node.Interface, nodeGET.PeerIDs)
  174. if err != nil {
  175. logger.Log(0, "failed metric collection for node", nodeCfg.Node.Name, err.Error())
  176. }
  177. metrics.Network = nodeCfg.Node.Network
  178. metrics.NodeName = nodeCfg.Node.Name
  179. metrics.NodeID = nodeCfg.Node.ID
  180. metrics.IsServer = "no"
  181. data, err := json.Marshal(metrics)
  182. if err != nil {
  183. logger.Log(0, "something went wrong when marshalling metrics data for node", nodeCfg.Node.Name, err.Error())
  184. }
  185. if err = publish(nodeCfg, fmt.Sprintf("metrics/%s", nodeCfg.Node.ID), data, 1); err != nil {
  186. logger.Log(0, "error occurred during publishing of metrics on node", nodeCfg.Node.Name, err.Error())
  187. logger.Log(0, "aggregating metrics locally until broker connection re-established")
  188. val, ok := metricsCache.Load(nodeCfg.Node.ID)
  189. if !ok {
  190. metricsCache.Store(nodeCfg.Node.ID, data)
  191. } else {
  192. var oldMetrics models.Metrics
  193. err = json.Unmarshal(val.([]byte), &oldMetrics)
  194. if err == nil {
  195. for k := range oldMetrics.Connectivity {
  196. currentMetric := metrics.Connectivity[k]
  197. if currentMetric.Latency == 0 {
  198. currentMetric.Latency = oldMetrics.Connectivity[k].Latency
  199. }
  200. currentMetric.Uptime += oldMetrics.Connectivity[k].Uptime
  201. currentMetric.TotalTime += oldMetrics.Connectivity[k].TotalTime
  202. metrics.Connectivity[k] = currentMetric
  203. }
  204. newData, err := json.Marshal(metrics)
  205. if err == nil {
  206. metricsCache.Store(nodeCfg.Node.ID, newData)
  207. }
  208. }
  209. }
  210. } else {
  211. metricsCache.Delete(nodeCfg.Node.ID)
  212. logger.Log(0, "published metrics for node", nodeCfg.Node.Name)
  213. }
  214. }
  215. // node cfg is required in order to fetch the traffic keys of that node for encryption
  216. func publish(nodeCfg *config.ClientConfig, dest string, msg []byte, qos byte) error {
  217. // setup the keys
  218. trafficPrivKey, err := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  219. if err != nil {
  220. return err
  221. }
  222. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  223. if err != nil {
  224. return err
  225. }
  226. encrypted, err := ncutils.Chunk(msg, serverPubKey, trafficPrivKey)
  227. if err != nil {
  228. return err
  229. }
  230. if mqclient == nil {
  231. return errors.New("unable to publish ... no mqclient")
  232. }
  233. if token := mqclient.Publish(dest, qos, false, encrypted); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
  234. logger.Log(0, "could not connect to broker at "+nodeCfg.Server.Server+":"+nodeCfg.Server.MQPort)
  235. var err error
  236. if token.Error() == nil {
  237. err = errors.New("connection timeout")
  238. } else {
  239. err = token.Error()
  240. }
  241. if err != nil {
  242. return err
  243. }
  244. }
  245. return nil
  246. }
  247. func checkBroker(broker string, port string) error {
  248. if broker == "" {
  249. return errors.New("error: broker address is blank")
  250. }
  251. if port == "" {
  252. return errors.New("error: broker port is blank")
  253. }
  254. _, err := net.LookupIP(broker)
  255. if err != nil {
  256. return errors.New("nslookup failed for broker ... check dns records")
  257. }
  258. pinger := ping.NewTCPing()
  259. intPort, err := strconv.Atoi(port)
  260. if err != nil {
  261. logger.Log(1, "error converting port to int: "+err.Error())
  262. }
  263. pinger.SetTarget(&ping.Target{
  264. Protocol: ping.TCP,
  265. Host: broker,
  266. Port: intPort,
  267. Counter: 3,
  268. Interval: 1 * time.Second,
  269. Timeout: 2 * time.Second,
  270. })
  271. pingerDone := pinger.Start()
  272. <-pingerDone
  273. if pinger.Result().SuccessCounter == 0 {
  274. return errors.New("unable to connect to broker port ... check netmaker server and firewalls")
  275. }
  276. return nil
  277. }