routeros-scripts/check-health

132 lines
5.4 KiB
Text
Raw Normal View History

#!rsc by RouterOS
# RouterOS script: check-health
2022-01-01 20:38:15 +00:00
# Copyright (c) 2019-2022 Christian Hesse <mail@eworm.de>
# https://git.eworm.de/cgit/routeros-scripts/about/COPYING.md
#
# check for RouterOS health state
2020-03-27 20:41:40 +00:00
# https://git.eworm.de/cgit/routeros-scripts/about/doc/check-health.md
:local 0 "check-health";
:global GlobalFunctionsReady;
:while ($GlobalFunctionsReady != true) do={ :delay 500ms; }
:global CheckHealthLast;
:global CheckHealthTemperature;
:global CheckHealthTemperatureDeviation;
:global CheckHealthTemperatureNotified;
:global CheckHealthVoltageLow;
:global CheckHealthVoltagePercent;
:global Identity;
:global IfThenElse;
:global LogPrintExit2;
2021-08-27 09:21:56 +00:00
:global ScriptLock;
2021-04-27 18:42:42 +00:00
:global SendNotification2;
:global SymbolForNotification;
:local TempToNum do={
:global CharacterReplace;
:local T [ :toarray [ $CharacterReplace $1 "." "," ] ];
:return ($T->0 * 10 + $T->1);
}
2022-05-10 12:43:04 +00:00
:if ([ :len [ /system/health/find ] ] = 0) do={
$LogPrintExit2 error $0 ("Your device does not provide any health values.") true;
}
:if ([ :typeof $CheckHealthLast ] != "array") do={
:set CheckHealthLast ({});
}
:if ([ :typeof $CheckHealthTemperatureNotified ] != "array") do={
:set CheckHealthTemperatureNotified ({});
}
2021-08-27 09:21:56 +00:00
$ScriptLock $0;
2022-05-10 12:43:04 +00:00
:foreach Voltage in=[ /system/health/find where type="V" ] do={
:local Name [ /system/health/get $Voltage name ];
:local Value [ /system/health/get $Voltage value ];
:if ([ :typeof ($CheckHealthLast->$Name) ] != "nothing") do={
:local NumCurr [ $TempToNum $Value ];
:local NumLast [ $TempToNum ($CheckHealthLast->$Name) ];
:if ($NumLast * (100 + $CheckHealthVoltagePercent) < $NumCurr * 100 || \
$NumLast * 100 > $NumCurr * (100 + $CheckHealthVoltagePercent)) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification ("high-voltage-sign,chart-" . [ $IfThenElse ($NumLast < \
$NumCurr) "in" "de" ] . "creasing") ] . "Health warning: " . $Name); \
2021-04-27 18:42:42 +00:00
message=("The " . $Name . " on " . $Identity . " jumped more than " . $CheckHealthVoltagePercent . "%.\n\n" . \
"old value: " . ($CheckHealthLast->$Name) . " V\n" . \
"new value: " . $Value . " V") });
} else={
:if ($NumCurr <= $CheckHealthVoltageLow && $NumLast > $CheckHealthVoltageLow) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "high-voltage-sign,chart-decreasing" ] . "Health warning: Low " . $Name); \
message=("The " . $Name . " on " . $Identity . " dropped to " . $Value . " V below hard limit.") });
}
:if ($NumCurr > $CheckHealthVoltageLow && $NumLast <= $CheckHealthVoltageLow) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "high-voltage-sign,chart-increasing" ] . "Health recovery: Low " . $Name); \
message=("The " . $Name . " on " . $Identity . " recovered to " . $Value . " V above hard limit.") });
}
}
}
:set ($CheckHealthLast->$Name) $Value;
}
2022-05-10 12:43:04 +00:00
:foreach PSU in=[ /system/health/find where name~"^psu.*-state\$" ] do={
:local Name [ /system/health/get $PSU name ];
:local Value [ /system/health/get $PSU value ];
:if ([ :typeof ($CheckHealthLast->$Name) ] != "nothing") do={
:if ($CheckHealthLast->$Name = "ok" && \
$Value != "ok") do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "cross-mark" ] . "Health warning: " . $Name); \
2021-04-27 18:42:42 +00:00
message=("The power supply unit '" . $Name . "' on " . $Identity . " failed!") });
}
:if ($CheckHealthLast->$Name != "ok" && \
$Value = "ok") do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "white-heavy-check-mark" ] . "Health recovery: " . $Name); \
2021-04-27 18:42:42 +00:00
message=("The power supply unit '" . $Name . "' on " . $Identity . " recovered!") });
}
}
:set ($CheckHealthLast->$Name) $Value;
}
2022-05-10 12:43:04 +00:00
:foreach Temperature in=[ /system/health/find where type="C" ] do={
:local Name [ /system/health/get $Temperature name ];
:local Value [ /system/health/get $Temperature value ];
:if ([ :typeof ($CheckHealthLast->$Name) ] != "nothing") do={
:if ([ :typeof ($CheckHealthTemperature->$Name) ] != "num" ) do={
$LogPrintExit2 info $0 ("No threshold given for " . $Name . ", assuming 50C.") false;
:set ($CheckHealthTemperature->$Name) 50;
}
2022-05-10 12:43:04 +00:00
:local Validate [ /system/health/get [ find where name=$Name ] value ];
:while ($Value != $Validate) do={
:set Value $Validate;
2022-05-10 12:43:04 +00:00
:set Validate [ /system/health/get [ find where name=$Name ] value ];
}
:if ($Value > $CheckHealthTemperature->$Name && \
$CheckHealthTemperatureNotified->$Name != true) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "fire" ] . "Health warning: " . $Name); \
2021-04-27 18:42:42 +00:00
message=("The " . $Name . " on " . $Identity . " is above threshold: " . \
$Value . "\C2\B0" . "C") });
:set ($CheckHealthTemperatureNotified->$Name) true;
}
:if ($Value <= ($CheckHealthTemperature->$Name - $CheckHealthTemperatureDeviation) && \
$CheckHealthTemperatureNotified->$Name = true) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "white-heavy-check-mark" ] . "Health recovery: " . $Name); \
2021-04-27 18:42:42 +00:00
message=("The " . $Name . " on " . $Identity . " dropped below threshold: " . \
$Value . "\C2\B0" . "C") });
:set ($CheckHealthTemperatureNotified->$Name) false;
}
}
:set ($CheckHealthLast->$Name) $Value;
}