2
0

zombie.go 5.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179
  1. package logic
  2. import (
  3. "context"
  4. "time"
  5. "github.com/google/uuid"
  6. "github.com/gravitl/netmaker/logger"
  7. "github.com/gravitl/netmaker/models"
  8. )
  9. const (
  10. // ZOMBIE_TIMEOUT - timeout in hours for checking zombie status
  11. ZOMBIE_TIMEOUT = 6
  12. // ZOMBIE_DELETE_TIME - timeout in minutes for zombie node deletion
  13. ZOMBIE_DELETE_TIME = 10
  14. )
  15. var (
  16. zombies []uuid.UUID
  17. hostZombies []uuid.UUID
  18. newZombie chan uuid.UUID = make(chan (uuid.UUID), 10)
  19. newHostZombie chan uuid.UUID = make(chan (uuid.UUID), 10)
  20. )
  21. // CheckZombies - checks if new node has same hostid as existing node
  22. // if so, existing node is added to zombie node quarantine list
  23. // also cleans up nodes past their expiration date
  24. func CheckZombies(newnode *models.Node) {
  25. nodes, err := GetNetworkNodes(newnode.Network)
  26. if err != nil {
  27. logger.Log(1, "Failed to retrieve network nodes", newnode.Network, err.Error())
  28. return
  29. }
  30. for _, node := range nodes {
  31. if node.ID == newnode.ID {
  32. //skip self
  33. continue
  34. }
  35. if node.HostID == newnode.HostID {
  36. logger.Log(0, "adding ", node.ID.String(), " to zombie list")
  37. newZombie <- node.ID
  38. }
  39. }
  40. }
  41. // checkForZombieHosts - checks if new host has the same macAddress as an existing host
  42. // if true, existing host is added to host zombie collection
  43. func checkForZombieHosts(h *models.Host) {
  44. hosts, err := GetAllHosts()
  45. if err != nil {
  46. logger.Log(3, "error retrieving all hosts", err.Error())
  47. }
  48. for _, existing := range hosts {
  49. if existing.ID == h.ID {
  50. //probably an unnecessary check as new host should not be in database yet, but just in case
  51. //skip self
  52. continue
  53. }
  54. if existing.MacAddress.String() == h.MacAddress.String() {
  55. //add to hostZombies
  56. newHostZombie <- existing.ID
  57. //add all nodes belonging to host to zombile list
  58. for _, node := range existing.Nodes {
  59. id, err := uuid.Parse(node)
  60. if err != nil {
  61. logger.Log(3, "error parsing uuid from host.Nodes", err.Error())
  62. continue
  63. }
  64. newHostZombie <- id
  65. }
  66. }
  67. }
  68. }
  69. // ManageZombies - goroutine which adds/removes/deletes nodes from the zombie node quarantine list
  70. func ManageZombies(ctx context.Context, peerUpdate chan *models.Node) {
  71. logger.Log(2, "Zombie management started")
  72. go InitializeZombies()
  73. go checkPendingRemovalNodes()
  74. // Zombie Nodes Cleanup Four Times a Day
  75. ticker := time.NewTicker(time.Hour * ZOMBIE_TIMEOUT)
  76. for {
  77. select {
  78. case <-ctx.Done():
  79. ticker.Stop()
  80. close(peerUpdate)
  81. return
  82. case id := <-newZombie:
  83. zombies = append(zombies, id)
  84. case id := <-newHostZombie:
  85. hostZombies = append(hostZombies, id)
  86. case <-ticker.C: // run this check 4 times a day
  87. logger.Log(3, "checking for zombie nodes")
  88. if len(zombies) > 0 {
  89. for i := len(zombies) - 1; i >= 0; i-- {
  90. node, err := GetNodeByID(zombies[i].String())
  91. if err != nil {
  92. logger.Log(1, "error retrieving zombie node", zombies[i].String(), err.Error())
  93. logger.Log(1, "deleting ", node.ID.String(), " from zombie list")
  94. zombies = append(zombies[:i], zombies[i+1:]...)
  95. continue
  96. }
  97. if time.Since(node.LastCheckIn) > time.Minute*ZOMBIE_DELETE_TIME {
  98. if err := DeleteNode(&node, true); err != nil {
  99. logger.Log(1, "error deleting zombie node", zombies[i].String(), err.Error())
  100. continue
  101. }
  102. node.PendingDelete = true
  103. node.Action = models.NODE_DELETE
  104. peerUpdate <- &node
  105. logger.Log(1, "deleting zombie node", node.ID.String())
  106. zombies = append(zombies[:i], zombies[i+1:]...)
  107. }
  108. }
  109. }
  110. if len(hostZombies) > 0 {
  111. logger.Log(3, "checking host zombies")
  112. for i := len(hostZombies) - 1; i >= 0; i-- {
  113. host, err := GetHost(hostZombies[i].String())
  114. if err != nil {
  115. logger.Log(1, "error retrieving zombie host", err.Error())
  116. if host != nil {
  117. logger.Log(1, "deleting ", host.ID.String(), " from zombie list")
  118. }
  119. hostZombies = append(hostZombies[:i], hostZombies[i+1:]...)
  120. continue
  121. }
  122. if len(host.Nodes) == 0 {
  123. if err := RemoveHost(host, true); err != nil {
  124. logger.Log(0, "error deleting zombie host", host.ID.String(), err.Error())
  125. }
  126. hostZombies = append(hostZombies[:i], hostZombies[i+1:]...)
  127. }
  128. }
  129. }
  130. }
  131. }
  132. }
  133. func checkPendingRemovalNodes() {
  134. nodes, _ := GetAllNodes()
  135. for _, node := range nodes {
  136. pendingDelete := node.PendingDelete || node.Action == models.NODE_DELETE
  137. if pendingDelete {
  138. DeleteNode(&node, true)
  139. }
  140. }
  141. }
  142. // InitializeZombies - populates the zombie quarantine list (should be called from initialization)
  143. func InitializeZombies() {
  144. nodes, err := GetAllNodes()
  145. if err != nil {
  146. logger.Log(1, "failed to retrieve nodes", err.Error())
  147. return
  148. }
  149. for _, node := range nodes {
  150. othernodes, err := GetNetworkNodes(node.Network)
  151. if err != nil {
  152. logger.Log(1, "failled to retrieve nodes for network", node.Network, err.Error())
  153. continue
  154. }
  155. for _, othernode := range othernodes {
  156. if node.ID == othernode.ID {
  157. continue
  158. }
  159. if node.HostID == othernode.HostID {
  160. if node.LastCheckIn.After(othernode.LastCheckIn) {
  161. newZombie <- othernode.ID
  162. logger.Log(1, "adding", othernode.ID.String(), "to zombie list")
  163. } else {
  164. newZombie <- node.ID
  165. logger.Log(1, "adding", node.ID.String(), "to zombie list")
  166. }
  167. }
  168. }
  169. }
  170. }