aboutsummaryrefslogtreecommitdiffstats
path: root/check-health
diff options
context:
space:
mode:
Diffstat (limited to 'check-health')
-rw-r--r--check-health72
1 files changed, 0 insertions, 72 deletions
diff --git a/check-health b/check-health
deleted file mode 100644
index 95a92d7..0000000
--- a/check-health
+++ /dev/null
@@ -1,72 +0,0 @@
-#!rsc
-# RouterOS script: check-health
-# Copyright (c) 2019-2020 Christian Hesse <mail@eworm.de>
-#
-# check for RouterOS health state
-
-:global CheckHealthLast;
-:global CheckHealthTemperature;
-:global CheckHealthVoltagePercent;
-: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={ "battery"; "psu1-voltage"; "psu2-voltage"; "voltage" } do={
- :if ([ :typeof ($CheckHealthLast->$Voltage) ] = "num" && \
- [ :typeof ($CheckHealthCurrent->$Voltage) ] = "num") do={
- :if ($CheckHealthLast->$Voltage * (100 + $CheckHealthVoltagePercent) / 100 < $CheckHealthCurrent->$Voltage || \
- $CheckHealthLast->$Voltage > $CheckHealthCurrent->$Voltage * (100 + $CheckHealthVoltagePercent) / 100) do={
- $SendNotification ("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 ("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;