mqpublish.go 10 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315
  1. package functions
  2. import (
  3. "context"
  4. "encoding/json"
  5. "errors"
  6. "fmt"
  7. "io"
  8. "net"
  9. "net/http"
  10. "os"
  11. "strconv"
  12. "sync"
  13. "time"
  14. "github.com/cloverstd/tcping/ping"
  15. "github.com/gravitl/netmaker/logger"
  16. "github.com/gravitl/netmaker/logic/metrics"
  17. "github.com/gravitl/netmaker/models"
  18. "github.com/gravitl/netmaker/netclient/auth"
  19. "github.com/gravitl/netmaker/netclient/config"
  20. "github.com/gravitl/netmaker/netclient/ncutils"
  21. "github.com/gravitl/netmaker/tls"
  22. )
  23. var metricsCache = new(sync.Map)
  24. // Checkin -- go routine that checks for public or local ip changes, publishes changes
  25. //
  26. // if there are no updates, simply "pings" the server as a checkin
  27. func Checkin(ctx context.Context, wg *sync.WaitGroup) {
  28. logger.Log(2, "starting checkin goroutine")
  29. defer wg.Done()
  30. currentRun := 0
  31. checkin(currentRun)
  32. ticker := time.NewTicker(time.Second * 60)
  33. defer ticker.Stop()
  34. for {
  35. select {
  36. case <-ctx.Done():
  37. logger.Log(0, "checkin routine closed")
  38. return
  39. //delay should be configuraable -> use cfg.Node.NetworkSettings.DefaultCheckInInterval ??
  40. case <-ticker.C:
  41. currentRun++
  42. checkin(currentRun)
  43. if currentRun >= 5 {
  44. currentRun = 0
  45. }
  46. }
  47. }
  48. }
  49. func checkin(currentRun int) {
  50. networks, _ := ncutils.GetSystemNetworks()
  51. logger.Log(3, "checkin with server(s) for all networks")
  52. for _, network := range networks {
  53. var nodeCfg config.ClientConfig
  54. nodeCfg.Network = network
  55. nodeCfg.ReadConfig()
  56. // check for nftables present if on Linux
  57. if ncutils.IsLinux() {
  58. if ncutils.IsNFTablesPresent() {
  59. nodeCfg.Node.FirewallInUse = models.FIREWALL_NFTABLES
  60. } else {
  61. nodeCfg.Node.FirewallInUse = models.FIREWALL_IPTABLES
  62. }
  63. } else {
  64. // defaults to iptables for now, may need another default for non-Linux OSes
  65. nodeCfg.Node.FirewallInUse = models.FIREWALL_IPTABLES
  66. }
  67. if nodeCfg.Node.IsStatic != "yes" {
  68. extIP, err := ncutils.GetPublicIP(nodeCfg.Server.API)
  69. if err != nil {
  70. logger.Log(1, "error encountered checking public ip addresses: ", err.Error())
  71. }
  72. if nodeCfg.Node.Endpoint != extIP && extIP != "" {
  73. logger.Log(1, "network:", nodeCfg.Node.Network, "endpoint has changed from ", nodeCfg.Node.Endpoint, " to ", extIP)
  74. nodeCfg.Node.Endpoint = extIP
  75. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  76. logger.Log(0, "network:", nodeCfg.Node.Network, "could not publish endpoint change")
  77. }
  78. }
  79. intIP, err := getPrivateAddr()
  80. if err != nil {
  81. logger.Log(1, "network:", nodeCfg.Node.Network, "error encountered checking private ip addresses: ", err.Error())
  82. }
  83. if nodeCfg.Node.LocalAddress != intIP && intIP != "" {
  84. logger.Log(1, "network:", nodeCfg.Node.Network, "local Address has changed from ", nodeCfg.Node.LocalAddress, " to ", intIP)
  85. nodeCfg.Node.LocalAddress = intIP
  86. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  87. logger.Log(0, "Network: ", nodeCfg.Node.Network, " could not publish local address change")
  88. }
  89. }
  90. _ = UpdateLocalListenPort(&nodeCfg)
  91. } else if nodeCfg.Node.IsLocal == "yes" && nodeCfg.Node.LocalRange != "" {
  92. localIP, err := ncutils.GetLocalIP(nodeCfg.Node.LocalRange)
  93. if err != nil {
  94. logger.Log(1, "network:", nodeCfg.Node.Network, "error encountered checking local ip addresses: ", err.Error())
  95. }
  96. if nodeCfg.Node.Endpoint != localIP && localIP != "" {
  97. logger.Log(1, "network:", nodeCfg.Node.Network, "endpoint has changed from "+nodeCfg.Node.Endpoint+" to ", localIP)
  98. nodeCfg.Node.Endpoint = localIP
  99. if err := PublishNodeUpdate(&nodeCfg); err != nil {
  100. logger.Log(0, "network:", nodeCfg.Node.Network, "could not publish localip change")
  101. }
  102. }
  103. }
  104. //check version
  105. if nodeCfg.Node.Version != ncutils.Version {
  106. nodeCfg.Node.Version = ncutils.Version
  107. config.Write(&nodeCfg, nodeCfg.Network)
  108. }
  109. Hello(&nodeCfg)
  110. checkCertExpiry(&nodeCfg)
  111. if currentRun >= 5 {
  112. logger.Log(0, "collecting metrics for node", nodeCfg.Node.Name)
  113. publishMetrics(&nodeCfg)
  114. }
  115. }
  116. }
  117. // PublishNodeUpdates -- saves node and pushes changes to broker
  118. func PublishNodeUpdate(nodeCfg *config.ClientConfig) error {
  119. if err := config.Write(nodeCfg, nodeCfg.Network); err != nil {
  120. return err
  121. }
  122. data, err := json.Marshal(nodeCfg.Node)
  123. if err != nil {
  124. return err
  125. }
  126. if err = publish(nodeCfg, fmt.Sprintf("update/%s", nodeCfg.Node.ID), data, 1); err != nil {
  127. return err
  128. }
  129. logger.Log(0, "network:", nodeCfg.Node.Network, "sent a node update to server for node", nodeCfg.Node.Name, ", ", nodeCfg.Node.ID)
  130. return nil
  131. }
  132. // Hello -- ping the broker to let server know node it's alive and well
  133. func Hello(nodeCfg *config.ClientConfig) {
  134. var checkin models.NodeCheckin
  135. checkin.Version = ncutils.Version
  136. checkin.Connected = nodeCfg.Node.Connected
  137. data, err := json.Marshal(checkin)
  138. if err != nil {
  139. logger.Log(0, "unable to marshal checkin data", err.Error())
  140. return
  141. }
  142. if err := publish(nodeCfg, fmt.Sprintf("ping/%s", nodeCfg.Node.ID), data, 0); err != nil {
  143. logger.Log(0, fmt.Sprintf("Network: %s error publishing ping, %v", nodeCfg.Node.Network, err))
  144. logger.Log(0, "running pull on "+nodeCfg.Node.Network+" to reconnect")
  145. _, err := Pull(nodeCfg.Node.Network, true)
  146. if err != nil {
  147. logger.Log(0, "could not run pull on "+nodeCfg.Node.Network+", error: "+err.Error())
  148. }
  149. } else {
  150. logger.Log(3, "checkin for", nodeCfg.Network, "complete")
  151. }
  152. }
  153. // publishMetrics - publishes the metrics of a given nodecfg
  154. func publishMetrics(nodeCfg *config.ClientConfig) {
  155. token, err := Authenticate(nodeCfg)
  156. if err != nil {
  157. logger.Log(1, "failed to authenticate when publishing metrics", err.Error())
  158. return
  159. }
  160. url := fmt.Sprintf("https://%s/api/nodes/%s/%s", nodeCfg.Server.API, nodeCfg.Network, nodeCfg.Node.ID)
  161. response, err := API("", http.MethodGet, url, token)
  162. if err != nil {
  163. logger.Log(1, "failed to read from server during metrics publish", err.Error())
  164. return
  165. }
  166. if response.StatusCode != http.StatusOK {
  167. bytes, err := io.ReadAll(response.Body)
  168. if err != nil {
  169. fmt.Println(err)
  170. }
  171. logger.Log(0, fmt.Sprintf("%s %s", string(bytes), err.Error()))
  172. return
  173. }
  174. defer response.Body.Close()
  175. var nodeGET models.NodeGet
  176. if err := json.NewDecoder(response.Body).Decode(&nodeGET); err != nil {
  177. logger.Log(0, "failed to decode node when running metrics update", err.Error())
  178. return
  179. }
  180. metrics, err := metrics.Collect(nodeCfg.Node.Interface, nodeGET.PeerIDs)
  181. if err != nil {
  182. logger.Log(0, "failed metric collection for node", nodeCfg.Node.Name, err.Error())
  183. }
  184. metrics.Network = nodeCfg.Node.Network
  185. metrics.NodeName = nodeCfg.Node.Name
  186. metrics.NodeID = nodeCfg.Node.ID
  187. metrics.IsServer = "no"
  188. data, err := json.Marshal(metrics)
  189. if err != nil {
  190. logger.Log(0, "something went wrong when marshalling metrics data for node", nodeCfg.Node.Name, err.Error())
  191. }
  192. if err = publish(nodeCfg, fmt.Sprintf("metrics/%s", nodeCfg.Node.ID), data, 1); err != nil {
  193. logger.Log(0, "error occurred during publishing of metrics on node", nodeCfg.Node.Name, err.Error())
  194. logger.Log(0, "aggregating metrics locally until broker connection re-established")
  195. val, ok := metricsCache.Load(nodeCfg.Node.ID)
  196. if !ok {
  197. metricsCache.Store(nodeCfg.Node.ID, data)
  198. } else {
  199. var oldMetrics models.Metrics
  200. err = json.Unmarshal(val.([]byte), &oldMetrics)
  201. if err == nil {
  202. for k := range oldMetrics.Connectivity {
  203. currentMetric := metrics.Connectivity[k]
  204. if currentMetric.Latency == 0 {
  205. currentMetric.Latency = oldMetrics.Connectivity[k].Latency
  206. }
  207. currentMetric.Uptime += oldMetrics.Connectivity[k].Uptime
  208. currentMetric.TotalTime += oldMetrics.Connectivity[k].TotalTime
  209. metrics.Connectivity[k] = currentMetric
  210. }
  211. newData, err := json.Marshal(metrics)
  212. if err == nil {
  213. metricsCache.Store(nodeCfg.Node.ID, newData)
  214. }
  215. }
  216. }
  217. } else {
  218. metricsCache.Delete(nodeCfg.Node.ID)
  219. logger.Log(0, "published metrics for node", nodeCfg.Node.Name)
  220. }
  221. }
  222. // node cfg is required in order to fetch the traffic keys of that node for encryption
  223. func publish(nodeCfg *config.ClientConfig, dest string, msg []byte, qos byte) error {
  224. // setup the keys
  225. trafficPrivKey, err := auth.RetrieveTrafficKey(nodeCfg.Node.Network)
  226. if err != nil {
  227. return err
  228. }
  229. serverPubKey, err := ncutils.ConvertBytesToKey(nodeCfg.Node.TrafficKeys.Server)
  230. if err != nil {
  231. return err
  232. }
  233. encrypted, err := ncutils.Chunk(msg, serverPubKey, trafficPrivKey)
  234. if err != nil {
  235. return err
  236. }
  237. if mqclient == nil {
  238. return errors.New("unable to publish ... no mqclient")
  239. }
  240. if token := mqclient.Publish(dest, qos, false, encrypted); !token.WaitTimeout(30*time.Second) || token.Error() != nil {
  241. logger.Log(0, "could not connect to broker at "+nodeCfg.Server.Server+":"+nodeCfg.Server.MQPort)
  242. var err error
  243. if token.Error() == nil {
  244. err = errors.New("connection timeout")
  245. } else {
  246. err = token.Error()
  247. }
  248. if err != nil {
  249. return err
  250. }
  251. }
  252. return nil
  253. }
  254. func checkCertExpiry(cfg *config.ClientConfig) error {
  255. cert, err := tls.ReadCertFromFile(ncutils.GetNetclientServerPath(cfg.Server.Server) + ncutils.GetSeparator() + "client.pem")
  256. //if cert doesn't exist or will expire within 10 days
  257. if errors.Is(err, os.ErrNotExist) || cert.NotAfter.Before(time.Now().Add(time.Hour*24*10)) {
  258. key, err := tls.ReadKeyFromFile(ncutils.GetNetclientPath() + ncutils.GetSeparator() + "client.key")
  259. if err != nil {
  260. return err
  261. }
  262. return RegisterWithServer(key, cfg)
  263. }
  264. if err != nil {
  265. return err
  266. }
  267. return nil
  268. }
  269. func checkBroker(broker string, port string) error {
  270. if broker == "" {
  271. return errors.New("error: broker address is blank")
  272. }
  273. if port == "" {
  274. return errors.New("error: broker port is blank")
  275. }
  276. _, err := net.LookupIP(broker)
  277. if err != nil {
  278. return errors.New("nslookup failed for broker ... check dns records")
  279. }
  280. pinger := ping.NewTCPing()
  281. intPort, err := strconv.Atoi(port)
  282. if err != nil {
  283. logger.Log(1, "error converting port to int: "+err.Error())
  284. }
  285. pinger.SetTarget(&ping.Target{
  286. Protocol: ping.TCP,
  287. Host: broker,
  288. Port: intPort,
  289. Counter: 3,
  290. Interval: 1 * time.Second,
  291. Timeout: 2 * time.Second,
  292. })
  293. pingerDone := pinger.Start()
  294. <-pingerDone
  295. if pinger.Result().SuccessCounter == 0 {
  296. return errors.New("unable to connect to broker port ... check netmaker server and firewalls")
  297. }
  298. return nil
  299. }