routeros-scripts/check-health
Christian Hesse 6d9eb99e08 check-health: add deviation on temperature recovery threshold
This helps against notification flooding.
2020-10-16 22:58:14 +02:00

88 lines
4 KiB
Plaintext

#!rsc by RouterOS
# RouterOS script: check-health
# Copyright (c) 2019-2020 Christian Hesse <mail@eworm.de>
# https://git.eworm.de/cgit/routeros-scripts/about/COPYING.md
#
# check for RouterOS health state
# https://git.eworm.de/cgit/routeros-scripts/about/doc/check-health.md
:global CheckHealthLast;
:global CheckHealthTemperature;
:global CheckHealthTemperatureDeviation;
:global CheckHealthTemperatureNotified;
:global CheckHealthVoltagePercent;
:global Identity;
:global LogPrintExit;
:global SendNotification;
:global SymbolForNotification;
:local FormatVoltage do={
:local Voltage [ :tonum $1 ];
:return (($Voltage / 10) . "." . ($Voltage % ($Voltage / 10 * 10)) . "V");
}
:local CheckHealthCurrent [ / system health get ];
:if ([ :len $CheckHealthCurrent ] = 0) do={
$LogPrintExit error ("Your device does not provide any health values.") true;
}
:if ([ :typeof $CheckHealthTemperatureNotified ] != "array") do={
:set CheckHealthTemperatureNotified [ :toarray "" ];
}
:foreach Voltage in={ "battery"; "psu1-voltage"; "psu2-voltage"; "voltage" } do={
:if ([ :typeof ($CheckHealthLast->$Voltage) ] = "num" && \
[ :typeof ($CheckHealthCurrent->$Voltage) ] = "num") do={
:if ($CheckHealthLast->$Voltage * (100 + $CheckHealthVoltagePercent) < $CheckHealthCurrent->$Voltage * 100 || \
$CheckHealthLast->$Voltage * 100 > $CheckHealthCurrent->$Voltage * (100 + $CheckHealthVoltagePercent)) do={
$SendNotification ([ $SymbolForNotification "high-voltage-sign" ] . "Health warning: " . $Voltage) \
("The " . $Voltage . " on " . $Identity . " jumped more than " . $CheckHealthVoltagePercent . "%.\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 ([ $SymbolForNotification "cross-mark" ] . "Health warning: " . $PSU . " state") \
("The power supply unit '" . $PSU . "' on " . $Identity . " failed!");
}
:if ($CheckHealthLast->($PSU . "-state") != "ok" && \
$CheckHealthCurrent->($PSU . "-state") = "ok") do={
$SendNotification ([ $SymbolForNotification "white-heavy-check-mark" ] . "Health recovery: " . $PSU . " state") \
("The power supply unit '" . $PSU . "' on " . $Identity . " recovered!");
}
}
}
:foreach Temperature in={ "temperature"; "cpu-temperature"; "board-temperature1"; "board-temperature2" } do={
:if ([ :typeof ($CheckHealthCurrent->$Temperature) ] = "num") do={
:if ([ :typeof ($CheckHealthTemperature->$Temperature) ] != "num" ) do={
$LogPrintExit warning ("No threshold given for " . $Temperature . ", assuming 50C.") false;
:set ($CheckHealthTemperature->$Temperature) 50;
}
:if ($CheckHealthCurrent->$Temperature > $CheckHealthTemperature->$Temperature && \
$CheckHealthTemperatureNotified->$Temperature != true) do={
$SendNotification ([ $SymbolForNotification "fire" ] . "Health warning: " . $Temperature) \
("The " . $Temperature . " on " . $Identity . " is above threshold: " . \
$CheckHealthCurrent->$Temperature . "\C2\B0" . "C");
:set ($CheckHealthTemperatureNotified->$Temperature) true;
}
:if ($CheckHealthCurrent->$Temperature <= ($CheckHealthTemperature->$Temperature - $CheckHealthTemperatureDeviation) && \
$CheckHealthTemperatureNotified->$Temperature = true) do={
$SendNotification ([ $SymbolForNotification "white-heavy-check-mark" ] . "Health recovery: " . $Temperature) \
("The " . $Temperature . " on " . $Identity . " dropped below threshold: " . \
$CheckHealthCurrent->$Temperature . "\C2\B0" . "C");
:set ($CheckHealthTemperatureNotified->$Temperature) false;
}
}
}
:set CheckHealthLast $CheckHealthCurrent;