| 1 | |
| 2 | # you can disable an alarm notification by setting the 'to' line to: silent |
| 3 | |
| 4 | # ----------------------------------------------------------------------------- |
| 5 | # net traffic overflow |
| 6 | |
| 7 | template: interface_speed |
| 8 | on: net.net |
| 9 | class: Latency |
| 10 | type: System |
| 11 | component: Network |
| 12 | calc: ( $nic_speed_max > 0 ) ? ( $nic_speed_max / 1000) : ( nan ) |
| 13 | units: Mbit |
| 14 | every: 10s |
| 15 | info: Network interface ${label:device} current speed |
| 16 | |
| 17 | template: 1m_received_traffic_overflow |
| 18 | on: net.net |
| 19 | class: Workload |
| 20 | type: System |
| 21 | component: Network |
| 22 | host labels: _os=linux windows |
| 23 | lookup: average -1m unaligned absolute of received |
| 24 | calc: ($interface_speed > 0) ? ($this * 100 / ($interface_speed * 1000)) : ( nan ) |
| 25 | units: % |
| 26 | every: 10s |
| 27 | warn: $this > (($status >= $WARNING) ? (85) : (90)) |
| 28 | delay: up 1m down 1m multiplier 1.5 max 1h |
| 29 | summary: System network interface ${label:device} inbound utilization |
| 30 | info: Average inbound utilization for the network interface ${label:device} over the last minute |
| 31 | to: silent |
| 32 | |
| 33 | template: 1m_sent_traffic_overflow |
| 34 | on: net.net |
| 35 | class: Workload |
| 36 | type: System |
| 37 | component: Network |
| 38 | host labels: _os=linux windows |
| 39 | lookup: average -1m unaligned absolute of sent |
| 40 | calc: ($interface_speed > 0) ? ($this * 100 / ($interface_speed * 1000)) : ( nan ) |
| 41 | units: % |
| 42 | every: 10s |
| 43 | warn: $this > (($status >= $WARNING) ? (85) : (90)) |
| 44 | delay: up 1m down 1m multiplier 1.5 max 1h |
| 45 | summary: System network interface ${label:device} outbound utilization |
| 46 | info: Average outbound utilization for the network interface ${label:device} over the last minute |
| 47 | to: silent |
| 48 | |
| 49 | # ----------------------------------------------------------------------------- |
| 50 | # dropped packets |
| 51 | |
| 52 | # check if an interface is dropping packets |
| 53 | # the alarm is checked every 1 minute |
| 54 | # and examines the last 10 minutes of data |
| 55 | # |
| 56 | # it is possible to have expected packet drops on an interface for some network configurations |
| 57 | # look at the Monitoring Network Interfaces section in the proc.plugin documentation for more information |
| 58 | |
| 59 | template: net_interface_inbound_packets |
| 60 | on: net.packets |
| 61 | class: Workload |
| 62 | type: System |
| 63 | component: Network |
| 64 | lookup: sum -10m unaligned absolute of received |
| 65 | units: packets |
| 66 | every: 1m |
| 67 | summary: Network interface ${label:device} received packets |
| 68 | info: Received packets for the network interface ${label:device} in the last 10 minutes |
| 69 | |
| 70 | template: net_interface_outbound_packets |
| 71 | on: net.packets |
| 72 | class: Workload |
| 73 | type: System |
| 74 | component: Network |
| 75 | lookup: sum -10m unaligned absolute of sent |
| 76 | units: packets |
| 77 | every: 1m |
| 78 | summary: Network interface ${label:device} sent packets |
| 79 | info: Sent packets for the network interface ${label:device} in the last 10 minutes |
| 80 | |
| 81 | template: inbound_packets_dropped_ratio |
| 82 | on: net.drops |
| 83 | class: Errors |
| 84 | type: System |
| 85 | component: Network |
| 86 | chart labels: device=!wl* * |
| 87 | lookup: sum -10m unaligned absolute of inbound |
| 88 | calc: (($net_interface_inbound_packets > 10000) ? ($this * 100 / $net_interface_inbound_packets) : (0)) |
| 89 | units: % |
| 90 | every: 1m |
| 91 | warn: $this >= 2 |
| 92 | delay: up 1m down 1h multiplier 1.5 max 2h |
| 93 | summary: System network interface ${label:device} inbound drops |
| 94 | info: Ratio of inbound dropped packets for the network interface ${label:device} over the last 10 minutes |
| 95 | to: silent |
| 96 | |
| 97 | template: outbound_packets_dropped_ratio |
| 98 | on: net.drops |
| 99 | class: Errors |
| 100 | type: System |
| 101 | component: Network |
| 102 | chart labels: device=!wl* * |
| 103 | lookup: sum -10m unaligned absolute of outbound |
| 104 | calc: (($net_interface_outbound_packets > 1000) ? ($this * 100 / $net_interface_outbound_packets) : (0)) |
| 105 | units: % |
| 106 | every: 1m |
| 107 | warn: $this >= 2 |
| 108 | delay: up 1m down 1h multiplier 1.5 max 2h |
| 109 | summary: System network interface ${label:device} outbound drops |
| 110 | info: Ratio of outbound dropped packets for the network interface ${label:device} over the last 10 minutes |
| 111 | to: silent |
| 112 | |
| 113 | template: wifi_inbound_packets_dropped_ratio |
| 114 | on: net.drops |
| 115 | class: Errors |
| 116 | type: System |
| 117 | component: Network |
| 118 | host labels: _os=linux |
| 119 | chart labels: device=wl* |
| 120 | lookup: sum -10m unaligned absolute of received |
| 121 | calc: (($net_interface_inbound_packets > 10000) ? ($this * 100 / $net_interface_inbound_packets) : (0)) |
| 122 | units: % |
| 123 | every: 1m |
| 124 | warn: $this >= 10 |
| 125 | delay: up 1m down 1h multiplier 1.5 max 2h |
| 126 | summary: System network interface ${label:device} inbound drops ratio |
| 127 | info: Ratio of inbound dropped packets for the network interface ${label:device} over the last 10 minutes |
| 128 | to: silent |
| 129 | |
| 130 | template: wifi_outbound_packets_dropped_ratio |
| 131 | on: net.drops |
| 132 | class: Errors |
| 133 | type: System |
| 134 | component: Network |
| 135 | host labels: _os=linux |
| 136 | chart labels: device=wl* |
| 137 | lookup: sum -10m unaligned absolute of sent |
| 138 | calc: (($net_interface_outbound_packets > 1000) ? ($this * 100 / $net_interface_outbound_packets) : (0)) |
| 139 | units: % |
| 140 | every: 1m |
| 141 | warn: $this >= 10 |
| 142 | delay: up 1m down 1h multiplier 1.5 max 2h |
| 143 | summary: System network interface ${label:device} outbound drops ratio |
| 144 | info: Ratio of outbound dropped packets for the network interface ${label:device} over the last 10 minutes |
| 145 | to: silent |
| 146 | |
| 147 | # ----------------------------------------------------------------------------- |
| 148 | # interface errors |
| 149 | |
| 150 | template: interface_inbound_errors |
| 151 | on: net.errors |
| 152 | class: Errors |
| 153 | type: System |
| 154 | component: Network |
| 155 | host labels: _os=freebsd |
| 156 | lookup: sum -10m unaligned absolute of inbound |
| 157 | units: errors |
| 158 | every: 1m |
| 159 | warn: $this >= 5 |
| 160 | delay: down 1h multiplier 1.5 max 2h |
| 161 | summary: System network interface ${label:device} inbound errors |
| 162 | info: Number of inbound errors for the network interface ${label:device} in the last 10 minutes |
| 163 | to: silent |
| 164 | |
| 165 | template: interface_outbound_errors |
| 166 | on: net.errors |
| 167 | class: Errors |
| 168 | type: System |
| 169 | component: Network |
| 170 | host labels: _os=freebsd |
| 171 | lookup: sum -10m unaligned absolute of outbound |
| 172 | units: errors |
| 173 | every: 1m |
| 174 | warn: $this >= 5 |
| 175 | delay: down 1h multiplier 1.5 max 2h |
| 176 | summary: System network interface ${label:device} outbound errors |
| 177 | info: Number of outbound errors for the network interface ${label:device} in the last 10 minutes |
| 178 | to: silent |
| 179 | |
| 180 | # ----------------------------------------------------------------------------- |
| 181 | # FIFO errors |
| 182 | |
| 183 | # check if an interface is having FIFO |
| 184 | # buffer errors |
| 185 | # the alarm is checked every 1 minute |
| 186 | # and examines the last 10 minutes of data |
| 187 | |
| 188 | template: 10min_fifo_errors |
| 189 | on: net.fifo |
| 190 | class: Errors |
| 191 | type: System |
| 192 | component: Network |
| 193 | host labels: _os=linux |
| 194 | lookup: sum -10m unaligned absolute |
| 195 | units: errors |
| 196 | every: 1m |
| 197 | warn: $this > 0 |
| 198 | delay: down 1h multiplier 1.5 max 2h |
| 199 | summary: System network interface ${label:device} FIFO errors |
| 200 | info: Number of FIFO errors for the network interface ${label:device} in the last 10 minutes |
| 201 | to: silent |
| 202 | |
| 203 | # ----------------------------------------------------------------------------- |
| 204 | # check for packet storms |
| 205 | |
| 206 | # 1. calculate the rate packets are received in 1m: 1m_received_packets_rate |
| 207 | # 2. do the same for the last 10s |
| 208 | # 3. raise an alarm if the later is 10x or 20x the first |
| 209 | # we assume the minimum packet storm should at least have |
| 210 | # 10000 packets/s, average of the last 10 seconds |
| 211 | |
| 212 | template: 1m_received_packets_rate |
| 213 | on: net.packets |
| 214 | class: Workload |
| 215 | type: System |
| 216 | component: Network |
| 217 | lookup: average -1m unaligned of received |
| 218 | units: packets |
| 219 | every: 10s |
| 220 | info: Average number of packets received by the network interface ${label:device} over the last minute |
| 221 | |
| 222 | template: 10s_received_packets_storm |
| 223 | on: net.packets |
| 224 | class: Workload |
| 225 | type: System |
| 226 | component: Network |
| 227 | lookup: average -10s unaligned of received |
| 228 | calc: $this * 100 / (($1m_received_packets_rate < 1000)?(1000):($1m_received_packets_rate)) |
| 229 | every: 10s |
| 230 | units: % |
| 231 | warn: $this > (($status >= $WARNING)?(200):(5000)) |
| 232 | crit: $this > (($status == $CRITICAL)?(5000):(6000)) |
| 233 | options: no-clear-notification |
| 234 | summary: System network interface ${label:device} inbound packet storm |
| 235 | info: Ratio of average number of received packets for the network interface ${label:device} over the last 10 seconds, \ |
| 236 | compared to the rate over the last minute |
| 237 | to: silent |
| 238 | |
| 239 | # ----------------------------------------------------------------------------- |
| 240 | # output queue length |
| 241 | |
| 242 | template: network_interface_output_queue_length |
| 243 | on: net.queue_length |
| 244 | class: Errors |
| 245 | type: System |
| 246 | component: Network |
| 247 | host labels: _os=windows |
| 248 | units: packets |
| 249 | every: 10s |
| 250 | warn: $length > 2 |
| 251 | delay: up 1m down 1m multiplier 1.5 max 1h |
| 252 | summary: System network interface ${label:device} output queue length |
| 253 | info: The Output Queue Length on interface ${label:device} should be zero, otherwise there are delays and bottlenecks. |
| 254 | to: silent |
| 255 |