#!rsc # RouterOS script: check-health # Copyright (c) 2019 Christian Hesse # # check for RouterOS health state :global CheckHealthLast; :global CheckHealthTemperature; :global Identity; :global SendNotification; :local FormatVoltage do={ :local Voltage [ :tonum $1 ]; :return (($Voltage / 10) . "." . ($Voltage % ($Voltage / 10 * 10)) . "V"); } :local CheckHealthCurrent [ / system health get ]; :foreach Voltage in={ "psu1-voltage"; "psu2-voltage"; "voltage" } do={ :if ([ :typeof ($CheckHealthLast->$Voltage) ] = "num" && \ [ :typeof ($CheckHealthCurrent->$Voltage) ] = "num") do={ :if ($CheckHealthLast->$Voltage * 115 / 100 < $CheckHealthCurrent->$Voltage || \ $CheckHealthLast->$Voltage * 100 / 115 > $CheckHealthCurrent->$Voltage) do={ $SendNotification ("Health warning: " . $Voltage) \ ("The " . $Voltage . " on " . $Identity . " jumped more than 15%.\n\n" . \ "old value: " . [ $FormatVoltage ($CheckHealthLast->$Voltage) ] . "\n" . \ "new value: " . [ $FormatVoltage ($CheckHealthCurrent->$Voltage) ]); } } } :foreach PSU in={ "psu1"; "psu2" } do={ :if ([ :typeof ($CheckHealthLast->($PSU . "-state")) ] = "str" && \ [ :typeof ($CheckHealthCurrent->($PSU . "-state")) ] = "str") do={ :if ($CheckHealthLast->($PSU . "-state") = "ok" && \ $CheckHealthCurrent->($PSU . "-state") != "ok") do={ $SendNotification ("Health warning: " . $PSU . " state") \ ("The power supply unit '" . $PSU . "' on " . $Identity . " failed!"); } :if ($CheckHealthLast->($PSU . "-state") != "ok" && \ $CheckHealthCurrent->($PSU . "-state") = "ok") do={ $SendNotification ("Health recovery: " . $PSU . " state") \ ("The power supply unit '" . $PSU . "' on " . $Identity . " recovered!"); } } } :foreach Temperature in={ "temperature"; "cpu-temperature"; "board-temperature1"; "board-temperature2" } do={ :if ([ :typeof ($CheckHealthLast->$Temperature) ] = "num" && \ [ :typeof ($CheckHealthCurrent->$Temperature) ] = "num") do={ :if ([ :typeof ($CheckHealthTemperature->$Temperature) ] != "num" ) do={ :log warning ("No threshold given for " . $Temperature . ", assuming 50C."); :set ($CheckHealthTemperature->$Temperature) 50; } :if ($CheckHealthLast->$Temperature <= $CheckHealthTemperature->$Temperature && \ $CheckHealthCurrent->$Temperature > $CheckHealthTemperature->$Temperature) do={ $SendNotification ("Health warning: " . $Temperature) \ ("The " . $Temperature . " on " . $Identity . " is above threshold: " . \ $CheckHealthCurrent->$Temperature . "C"); } :if ($CheckHealthLast->$Temperature > $CheckHealthTemperature->$Temperature && \ $CheckHealthCurrent->$Temperature <= $CheckHealthTemperature->$Temperature) do={ $SendNotification ("Health recovery: " . $Temperature) \ ("The " . $Temperature . " on " . $Identity . " dropped below threshold: " . \ $CheckHealthCurrent->$Temperature . "C"); } } } :set CheckHealthLast $CheckHealthCurrent;