vsphere.conf 5.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174
  1. # you can disable an alarm notification by setting the 'to' line to: silent
  2. # -----------------------------------------------VM Specific------------------------------------------------------------
  3. # Memory
  4. template: vsphere_vm_mem_usage
  5. on: vsphere.vm_mem_usage_percentage
  6. class: Utilization
  7. type: Virtual Machine
  8. component: Memory
  9. hosts: *
  10. calc: $used
  11. units: %
  12. every: 20s
  13. warn: $this > (($status >= $WARNING) ? (80) : (90))
  14. crit: $this > (($status == $CRITICAL) ? (90) : (98))
  15. delay: down 15m multiplier 1.5 max 1h
  16. info: virtual machine memory utilization
  17. # -----------------------------------------------HOST Specific----------------------------------------------------------
  18. # Memory
  19. template: vsphere_host_mem_usage
  20. on: vsphere.host_mem_usage_percentage
  21. class: Utilization
  22. type: Virtual Machine
  23. component: Memory
  24. hosts: *
  25. calc: $used
  26. units: %
  27. every: 20s
  28. warn: $this > (($status >= $WARNING) ? (80) : (90))
  29. crit: $this > (($status == $CRITICAL) ? (90) : (98))
  30. delay: down 15m multiplier 1.5 max 1h
  31. info: host memory utilization
  32. # Network errors
  33. template: vsphere_inbound_packets_errors
  34. on: vsphere.net_errors_total
  35. class: Errors
  36. type: Virtual Machine
  37. component: Network
  38. hosts: *
  39. families: *
  40. lookup: sum -10m unaligned absolute match-names of rx
  41. units: packets
  42. every: 1m
  43. info: number of inbound errors for the network interface in the last 10 minutes
  44. template: vsphere_outbound_packets_errors
  45. on: vsphere.net_errors_total
  46. class: Errors
  47. type: Virtual Machine
  48. component: Network
  49. hosts: *
  50. families: *
  51. lookup: sum -10m unaligned absolute match-names of tx
  52. units: packets
  53. every: 1m
  54. info: number of outbound errors for the network interface in the last 10 minutes
  55. # Network errors ratio
  56. template: vsphere_inbound_packets_errors_ratio
  57. on: vsphere.net_packets_total
  58. class: Errors
  59. type: Virtual Machine
  60. component: Network
  61. hosts: *
  62. families: *
  63. lookup: sum -10m unaligned absolute match-names of rx
  64. calc: (($vsphere_inbound_packets_errors != nan AND $this > 1000) ? ($vsphere_inbound_packets_errors * 100 / $this) : (0))
  65. units: %
  66. every: 1m
  67. warn: $this >= 2
  68. delay: up 1m down 1h multiplier 1.5 max 2h
  69. info: ratio of inbound errors for the network interface over the last 10 minutes
  70. to: sysadmin
  71. template: vsphere_outbound_packets_errors_ratio
  72. on: vsphere.net_packets_total
  73. class: Errors
  74. type: Virtual Machine
  75. component: Network
  76. hosts: *
  77. families: *
  78. lookup: sum -10m unaligned absolute match-names of tx
  79. calc: (($vsphere_outbound_packets_errors != nan AND $this > 1000) ? ($vsphere_outbound_packets_errors * 100 / $this) : (0))
  80. units: %
  81. every: 1m
  82. warn: $this >= 2
  83. delay: up 1m down 1h multiplier 1.5 max 2h
  84. info: ratio of outbound errors for the network interface over the last 10 minutes
  85. to: sysadmin
  86. # -----------------------------------------------Common-------------------------------------------------------------------
  87. # CPU
  88. template: vsphere_cpu_usage
  89. on: vsphere.cpu_usage_total
  90. class: Utilization
  91. type: Virtual Machine
  92. component: CPU
  93. hosts: *
  94. lookup: average -10m unaligned match-names of used
  95. units: %
  96. every: 20s
  97. warn: $this > (($status >= $WARNING) ? (75) : (85))
  98. crit: $this > (($status == $CRITICAL) ? (85) : (95))
  99. delay: down 15m multiplier 1.5 max 1h
  100. info: average CPU utilization
  101. to: sysadmin
  102. # Network drops
  103. template: vsphere_inbound_packets_dropped
  104. on: vsphere.net_drops_total
  105. class: Errors
  106. type: Virtual Machine
  107. component: Network
  108. hosts: *
  109. families: *
  110. lookup: sum -10m unaligned absolute match-names of rx
  111. units: packets
  112. every: 1m
  113. info: number of inbound dropped packets for the network interface in the last 10 minutes
  114. template: vsphere_outbound_packets_dropped
  115. on: vsphere.net_drops_total
  116. class: Errors
  117. type: Virtual Machine
  118. component: Network
  119. hosts: *
  120. families: *
  121. lookup: sum -10m unaligned absolute match-names of tx
  122. units: packets
  123. every: 1m
  124. info: number of outbound dropped packets for the network interface in the last 10 minutes
  125. # Network drops ratio
  126. template: vsphere_inbound_packets_dropped_ratio
  127. on: vsphere.net_packets_total
  128. class: Errors
  129. type: Virtual Machine
  130. component: Network
  131. hosts: *
  132. families: *
  133. lookup: sum -10m unaligned absolute match-names of rx
  134. calc: (($vsphere_inbound_packets_dropped != nan AND $this > 1000) ? ($vsphere_inbound_packets_dropped * 100 / $this) : (0))
  135. units: %
  136. every: 1m
  137. warn: $this >= 2
  138. delay: up 1m down 1h multiplier 1.5 max 2h
  139. info: ratio of inbound dropped packets for the network interface over the last 10 minutes
  140. to: sysadmin
  141. template: vsphere_outbound_packets_dropped_ratio
  142. on: vsphere.net_packets_total
  143. class: Errors
  144. type: Virtual Machine
  145. component: Network
  146. hosts: *
  147. families: *
  148. lookup: sum -10m unaligned absolute match-names of tx
  149. calc: (($vsphere_outbound_packets_dropped != nan AND $this > 1000) ? ($vsphere_outbound_packets_dropped * 100 / $this) : (0))
  150. units: %
  151. every: 1m
  152. warn: $this >= 2
  153. delay: up 1m down 1h multiplier 1.5 max 2h
  154. info: ratio of outbound dropped packets for the network interface over the last 10 minutes
  155. to: sysadmin