mirror.MikroTik/eworm.routeros-scripts
1
0
Fork 0
mirror of https://git.eworm.de/cgit/routeros-scripts synced 2025-06-21 01:15:42 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions

check-health: move code into function

Browse source
This commit is contained in:
Christian Hesse 2024-03-04 13:48:00 +01:00
parent fc3fad5e87
commit 80180b432d
1 changed files with 147 additions and 142 deletions
Download patch file Download diff file Expand all files Collapse all files

289
check-health.rsc
View file

@ -8,165 +8,170 @@
# check for RouterOS health state # check for RouterOS health state
# https://git.eworm.de/cgit/routeros-scripts/about/doc/check-health.md # https://git.eworm.de/cgit/routeros-scripts/about/doc/check-health.md
:local 0 [ :jobname ];
:global GlobalFunctionsReady; :global GlobalFunctionsReady;
:while ($GlobalFunctionsReady != true) do={ :delay 500ms; } :while ($GlobalFunctionsReady != true) do={ :delay 500ms; }
:global CheckHealthCPUUtilization; :local Main do={
:global CheckHealthCPUUtilizationNotified; :local ScriptName [ :tostr $1 ];
:global CheckHealthLast;
:global CheckHealthRAMUtilizationNotified;
:global CheckHealthTemperature;
:global CheckHealthTemperatureDeviation;
:global CheckHealthTemperatureNotified;
:global CheckHealthVoltageLow;
:global CheckHealthVoltagePercent;
:global Identity;
:global FormatLine; :global CheckHealthCPUUtilization;
:global HumanReadableNum; :global CheckHealthCPUUtilizationNotified;
:global IfThenElse; :global CheckHealthLast;
:global LogPrintExit2; :global CheckHealthRAMUtilizationNotified;
:global ScriptLock; :global CheckHealthTemperature;
:global SendNotification2; :global CheckHealthTemperatureDeviation;
:global SymbolForNotification; :global CheckHealthTemperatureNotified;
:global CheckHealthVoltageLow;
:global CheckHealthVoltagePercent;
:global Identity;
:local TempToNum do={ :global FormatLine;
:global CharacterReplace; :global HumanReadableNum;
:local T [ :toarray [ $CharacterReplace $1 "." "," ] ]; :global IfThenElse;
:return ($T->0 * 10 + $T->1); :global LogPrintExit2;
} :global ScriptLock;
:global SendNotification2;
:global SymbolForNotification;
$ScriptLock $0; :local TempToNum do={
:global CharacterReplace;
:local T [ :toarray [ $CharacterReplace $1 "." "," ] ];
:return ($T->0 * 10 + $T->1);
}
:local Resource [ /system/resource/get ]; $ScriptLock $ScriptName;
:set CheckHealthCPUUtilization (($CheckHealthCPUUtilization * 4 + ($Resource->"cpu-load") * 10) / 5); :local Resource [ /system/resource/get ];
:if ($CheckHealthCPUUtilization > 750 && $CheckHealthCPUUtilizationNotified != true) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "abacus,chart-increasing" ] . "Health warning: CPU utilization"); \
message=("The average CPU utilization on " . $Identity . " is at " . ($CheckHealthCPUUtilization / 10) . "%!") });
:set CheckHealthCPUUtilizationNotified true;
}
:if ($CheckHealthCPUUtilization < 650 && $CheckHealthCPUUtilizationNotified = true) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "abacus,chart-decreasing" ] . "Health recovery: CPU utilization"); \
message=("The average CPU utilization on " . $Identity . " decreased to " . ($CheckHealthCPUUtilization / 10) . "%.") });
:set CheckHealthCPUUtilizationNotified false;
}
:local CheckHealthRAMUtilization (($Resource->"total-memory" - $Resource->"free-memory") * 100 / $Resource->"total-memory"); :set CheckHealthCPUUtilization (($CheckHealthCPUUtilization * 4 + ($Resource->"cpu-load") * 10) / 5);
:if ($CheckHealthRAMUtilization >=80 && $CheckHealthRAMUtilizationNotified != true) do={ :if ($CheckHealthCPUUtilization > 750 && $CheckHealthCPUUtilizationNotified != true) do={
$SendNotification2 ({ origin=$0; \ $SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "card-file-box,chart-increasing" ] . "Health warning: RAM utilization"); \ subject=([ $SymbolForNotification "abacus,chart-increasing" ] . "Health warning: CPU utilization"); \
message=("The RAM utilization on " . $Identity . " is at " . $CheckHealthRAMUtilization . "%!\n\n" . \ message=("The average CPU utilization on " . $Identity . " is at " . ($CheckHealthCPUUtilization / 10) . "%!") });
[ $FormatLine "total" ([ $HumanReadableNum ($Resource->"total-memory") 1024 ] . "iB") 8 ] . "\n" . \ :set CheckHealthCPUUtilizationNotified true;
[ $FormatLine "used" ([ $HumanReadableNum ($Resource->"total-memory" - $Resource->"free-memory") 1024 ] . "iB") 8 ] . "\n" . \ }
[ $FormatLine "free" ([ $HumanReadableNum ($Resource->"free-memory") 1024 ] . "iB") 8 ]) }); :if ($CheckHealthCPUUtilization < 650 && $CheckHealthCPUUtilizationNotified = true) do={
:set CheckHealthRAMUtilizationNotified true; $SendNotification2 ({ origin=$ScriptName; \
} subject=([ $SymbolForNotification "abacus,chart-decreasing" ] . "Health recovery: CPU utilization"); \
:if ($CheckHealthRAMUtilization < 70 && $CheckHealthRAMUtilizationNotified = true) do={ message=("The average CPU utilization on " . $Identity . " decreased to " . ($CheckHealthCPUUtilization / 10) . "%.") });
$SendNotification2 ({ origin=$0; \ :set CheckHealthCPUUtilizationNotified false;
subject=([ $SymbolForNotification "card-file-box,chart-decreasing" ] . "Health recovery: RAM utilization"); \ }
message=("The RAM utilization on " . $Identity . " decreased to " . $CheckHealthRAMUtilization . "%.") });
:set CheckHealthRAMUtilizationNotified false;
}
:if ([ :len [ /system/health/find ] ] = 0) do={ :local CheckHealthRAMUtilization (($Resource->"total-memory" - $Resource->"free-memory") * 100 / $Resource->"total-memory");
$LogPrintExit2 debug $0 ("Your device does not provide any health values.") true; :if ($CheckHealthRAMUtilization >=80 && $CheckHealthRAMUtilizationNotified != true) do={
} $SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "card-file-box,chart-increasing" ] . "Health warning: RAM utilization"); \
message=("The RAM utilization on " . $Identity . " is at " . $CheckHealthRAMUtilization . "%!\n\n" . \
[ $FormatLine "total" ([ $HumanReadableNum ($Resource->"total-memory") 1024 ] . "iB") 8 ] . "\n" . \
[ $FormatLine "used" ([ $HumanReadableNum ($Resource->"total-memory" - $Resource->"free-memory") 1024 ] . "iB") 8 ] . "\n" . \
[ $FormatLine "free" ([ $HumanReadableNum ($Resource->"free-memory") 1024 ] . "iB") 8 ]) });
:set CheckHealthRAMUtilizationNotified true;
}
:if ($CheckHealthRAMUtilization < 70 && $CheckHealthRAMUtilizationNotified = true) do={
$SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "card-file-box,chart-decreasing" ] . "Health recovery: RAM utilization"); \
message=("The RAM utilization on " . $Identity . " decreased to " . $CheckHealthRAMUtilization . "%.") });
:set CheckHealthRAMUtilizationNotified false;
}
:if ([ :typeof $CheckHealthLast ] != "array") do={ :if ([ :len [ /system/health/find ] ] = 0) do={
:set CheckHealthLast ({}); $LogPrintExit2 debug $ScriptName ("Your device does not provide any health values.") true;
} }
:if ([ :typeof $CheckHealthTemperatureNotified ] != "array") do={
:set CheckHealthTemperatureNotified ({}); :if ([ :typeof $CheckHealthLast ] != "array") do={
} :set CheckHealthLast ({});
}
:if ([ :typeof $CheckHealthTemperatureNotified ] != "array") do={
:set CheckHealthTemperatureNotified ({});
}
:foreach Voltage in=[ /system/health/find where type="V" ] do={ :foreach Voltage in=[ /system/health/find where type="V" ] do={
:local Name [ /system/health/get $Voltage name ]; :local Name [ /system/health/get $Voltage name ];
:local Value [ /system/health/get $Voltage value ]; :local Value [ /system/health/get $Voltage value ];
:if ([ :typeof ($CheckHealthLast->$Name) ] != "nothing") do={ :if ([ :typeof ($CheckHealthLast->$Name) ] != "nothing") do={
:local NumCurr [ $TempToNum $Value ]; :local NumCurr [ $TempToNum $Value ];
:local NumLast [ $TempToNum ($CheckHealthLast->$Name) ]; :local NumLast [ $TempToNum ($CheckHealthLast->$Name) ];
:if ($NumLast * (100 + $CheckHealthVoltagePercent) < $NumCurr * 100 || \ :if ($NumLast * (100 + $CheckHealthVoltagePercent) < $NumCurr * 100 || \
$NumLast * 100 > $NumCurr * (100 + $CheckHealthVoltagePercent)) do={ $NumLast * 100 > $NumCurr * (100 + $CheckHealthVoltagePercent)) do={
$SendNotification2 ({ origin=$0; \ $SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification ("high-voltage-sign,chart-" . [ $IfThenElse ($NumLast < \ subject=([ $SymbolForNotification ("high-voltage-sign,chart-" . [ $IfThenElse ($NumLast < \
$NumCurr) "in" "de" ] . "creasing") ] . "Health warning: " . $Name); \ $NumCurr) "in" "de" ] . "creasing") ] . "Health warning: " . $Name); \
message=("The " . $Name . " on " . $Identity . " jumped more than " . $CheckHealthVoltagePercent . "%.\n\n" . \ message=("The " . $Name . " on " . $Identity . " jumped more than " . $CheckHealthVoltagePercent . "%.\n\n" . \
[ $FormatLine "old value" ($CheckHealthLast->$Name . " V") 12 ] . "\n" . \ [ $FormatLine "old value" ($CheckHealthLast->$Name . " V") 12 ] . "\n" . \
[ $FormatLine "new value" ($Value . " V") 12 ]) }); [ $FormatLine "new value" ($Value . " V") 12 ]) });
} else={ } else={
:if ($NumCurr <= $CheckHealthVoltageLow && $NumLast > $CheckHealthVoltageLow) do={ :if ($NumCurr <= $CheckHealthVoltageLow && $NumLast > $CheckHealthVoltageLow) do={
$SendNotification2 ({ origin=$0; \ $SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "high-voltage-sign,chart-decreasing" ] . "Health warning: Low " . $Name); \ subject=([ $SymbolForNotification "high-voltage-sign,chart-decreasing" ] . "Health warning: Low " . $Name); \
message=("The " . $Name . " on " . $Identity . " dropped to " . $Value . " V below hard limit.") }); message=("The " . $Name . " on " . $Identity . " dropped to " . $Value . " V below hard limit.") });
} }
:if ($NumCurr > $CheckHealthVoltageLow && $NumLast <= $CheckHealthVoltageLow) do={ :if ($NumCurr > $CheckHealthVoltageLow && $NumLast <= $CheckHealthVoltageLow) do={
$SendNotification2 ({ origin=$0; \ $SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "high-voltage-sign,chart-increasing" ] . "Health recovery: Low " . $Name); \ subject=([ $SymbolForNotification "high-voltage-sign,chart-increasing" ] . "Health recovery: Low " . $Name); \
message=("The " . $Name . " on " . $Identity . " recovered to " . $Value . " V above hard limit.") }); message=("The " . $Name . " on " . $Identity . " recovered to " . $Value . " V above hard limit.") });
}
} }
} }
:set ($CheckHealthLast->$Name) $Value;
}
: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={
$SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "cross-mark" ] . "Health warning: " . $Name); \
message=("The power supply unit '" . $Name . "' on " . $Identity . " failed!") });
}
:if ($CheckHealthLast->$Name != "ok" && \
$Value = "ok") do={
$SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "white-heavy-check-mark" ] . "Health recovery: " . $Name); \
message=("The power supply unit '" . $Name . "' on " . $Identity . " recovered!") });
}
}
:set ($CheckHealthLast->$Name) $Value;
}
: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 $ScriptName ("No threshold given for " . $Name . ", assuming 50C.") false;
:set ($CheckHealthTemperature->$Name) 50;
}
:local Validate [ /system/health/get [ find where name=$Name ] value ];
:while ($Value != $Validate) do={
:set Value $Validate;
:set Validate [ /system/health/get [ find where name=$Name ] value ];
}
:if ($Value > $CheckHealthTemperature->$Name && \
$CheckHealthTemperatureNotified->$Name != true) do={
$SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "fire" ] . "Health warning: " . $Name); \
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={
$SendNotification2 ({ origin=$ScriptName; \
subject=([ $SymbolForNotification "white-heavy-check-mark" ] . "Health recovery: " . $Name); \
message=("The " . $Name . " on " . $Identity . " dropped below threshold: " . \
$Value . "\C2\B0" . "C") });
:set ($CheckHealthTemperatureNotified->$Name) false;
}
}
:set ($CheckHealthLast->$Name) $Value;
} }
:set ($CheckHealthLast->$Name) $Value;
} }
:foreach PSU in=[ /system/health/find where name~"^psu.*-state\$" ] do={ $Main [ :jobname ];
: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); \
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); \
message=("The power supply unit '" . $Name . "' on " . $Identity . " recovered!") });
}
}
:set ($CheckHealthLast->$Name) $Value;
}
: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->$Name) 50;
}
:local Validate [ /system/health/get [ find where name=$Name ] value ];
:while ($Value != $Validate) do={
:set Value $Validate;
:set Validate [ /system/health/get [ find where name=$Name ] value ];
}
:if ($Value > $CheckHealthTemperature->$Name && \
$CheckHealthTemperatureNotified->$Name != true) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "fire" ] . "Health warning: " . $Name); \
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); \
message=("The " . $Name . " on " . $Identity . " dropped below threshold: " . \
$Value . "\C2\B0" . "C") });
:set ($CheckHealthTemperatureNotified->$Name) false;
}
}
:set ($CheckHealthLast->$Name) $Value;
}