routeros-scripts/netwatch-notify
Christian Hesse 0c607a8f9c netwatch-notify: nest conditions
The logic here was right, but RouterOS runs the checks simultaneously.
This caused delays even if no resolving was needed.

Nesting the checks fixes this.

(cherry picked from commit 0b46c508dc)
2022-02-10 09:14:49 +01:00

155 lines
6.4 KiB
Plaintext

#!rsc by RouterOS
# RouterOS script: netwatch-notify
# Copyright (c) 2020-2022 Christian Hesse <mail@eworm.de>
# https://git.eworm.de/cgit/routeros-scripts/about/COPYING.md
#
# monitor netwatch and send notifications
# https://git.eworm.de/cgit/routeros-scripts/about/doc/netwatch-notify.md
:local 0 "netwatch-notify";
:global GlobalFunctionsReady;
:while ($GlobalFunctionsReady != true) do={ :delay 500ms; }
:global NetwatchNotify;
:global DNSIsResolving;
:global IfThenElse;
:global LogPrintExit2;
:global ParseKeyValueStore;
:global ScriptLock;
:global SendNotification2;
:global SymbolForNotification;
:local NetwatchNotifyHook do={
:local Name [ :tostr $1 ];
:local Type [ :tostr $2 ];
:local Hook [ :tostr $3 ];
:global LogPrintExit2;
:global ValidateSyntax;
:if ([ $ValidateSyntax $Hook ] = true) do={
:do {
[ :parse $Hook ];
} on-error={
$LogPrintExit2 warning $0 ("The " . $Type . "-hook for host " . $Name . " failed to run.") false;
:return ("The hook failed to run.");
}
} else={
$LogPrintExit2 warning $0 ("The " . $Type . "-hook for host " . $Name . " failed syntax validation.") false;
:return ("The hook failed syntax validation.");
}
$LogPrintExit2 info $0 ("Ran hook on host " . $Name . " " . $Type . ": " . $Hook) false;
:return ("Ran hook:\n" . $Hook);
}
$ScriptLock $0;
:if ([ / system resource get uptime ] < 5m) do={
$LogPrintExit2 info $0 ("System just booted, giving netwatch some time to settle.") true;
}
:if ([ :typeof $NetwatchNotify ] = "nothing") do={
:set NetwatchNotify [ :toarray "" ];
}
:foreach Host in=[ / tool netwatch find where comment~"notify" disabled=no ] do={
:local HostVal [ / tool netwatch get $Host ];
:local HostInfo [ $ParseKeyValueStore ($HostVal->"comment") ];
:if ($HostInfo->"notify" = true && $HostInfo->"disabled" != true) do={
:local HostName ($HostInfo->"hostname");
:local Metric { "count"=0; "notified"=false };
:if ([ :typeof ($NetwatchNotify->$HostName) ] = "array") do={
:set $Metric ($NetwatchNotify->$HostName);
}
:if ([ :typeof ($HostInfo->"resolve") ] = "str") do={
:if ([ $DNSIsResolving ] = true) do={
:do {
:local Resolve [ :resolve ($HostInfo->"resolve") ];
:if ($Resolve != $HostVal->"host") do={
$LogPrintExit2 info $0 ("Name '" . $HostInfo->"resolve" . [ $IfThenElse ($HostInfo->"resolve" != \
$HostInfo->"hostname") ("' for host '" . $HostInfo->"hostname") "" ] . \
"' resolves to different address " . $Resolve . ", updating.") false;
/ tool netwatch set host=$Resolve $Host;
:set ($Metric->"resolve-failed") false;
}
} on-error={
:if ($Metric->"resolve-failed" != true) do={
$LogPrintExit2 warning $0 ("Resolving name '" . $HostInfo->"resolve" . [ $IfThenElse ($HostInfo->"resolve" != \
$HostInfo->"hostname") ("' for host '" . $HostInfo->"hostname") "" ] . "' failed.") false;
:set ($Metric->"resolve-failed") true;
}
}
}
}
:if ($HostVal->"status" = "up") do={
:local Count ($Metric->"count");
:if ($Count > 0) do={
$LogPrintExit2 info $0 ("Host " . $HostName . " (" . $HostVal->"host" . ") is up.") false;
:set ($Metric->"count") 0;
}
:if ($Metric->"notified" = true) do={
:local Message ("Host " . $HostName . " (" . $HostVal->"host" . ") is up since " . $HostVal->"since" . ".\n" . \
"It was down for " . $Count . " checks since " . ($Metric->"since") . ".");
:if ([ :typeof ($HostInfo->"up-hook") ] = "str") do={
:set Message ($Message . "\n\n" . [ $NetwatchNotifyHook $HostName "up" ($HostInfo->"up-hook") ]);
}
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "white-heavy-check-mark" ] . "Netwatch Notify: " . $HostName . " up"); \
message=$Message });
}
:set ($Metric->"notified") false;
:set ($Metric->"parent") ($HostInfo->"parent");
:set ($Metric->"since");
} else={
:set ($Metric->"count") ($Metric->"count" + 1);
:set ($Metric->"parent") ($HostInfo->"parent");
:set ($Metric->"since") ($HostVal->"since");
:local Count [ $IfThenElse ([ :tonum ($HostInfo->"count") ] > 0) ($HostInfo->"count") 5 ];
:local Parent ($HostInfo->"parent");
:while ([ :len $Parent ] > 0) do={
:set Count ($Count + 1);
:set Parent ($NetwatchNotify->$Parent->"parent");
}
:set Parent ($HostInfo->"parent");
:local ParentNotified false;
:while ($ParentNotified = false && [ :len $Parent ] > 0) do={
:set ParentNotified [ $IfThenElse (($NetwatchNotify->$Parent->"notified") = true) true false ];
:if ($ParentNotified = false) do={
:set Parent ($NetwatchNotify->$Parent->"parent");
}
}
$LogPrintExit2 [ $IfThenElse ($HostInfo->"no-down-notification" != true) info debug ] $0 \
("Host " . $HostName . " (" . $HostVal->"host" . ") is down for " . $Metric->"count" . " checks, " . \
[ $IfThenElse ($ParentNotified = false) [ $IfThenElse ($Metric->"notified" = true) ("already notified.") \
($Count - $Metric->"count" . " to go.") ] ("parent host " . $Parent . " is down.") ]) false;
:if ((($Count * 2) - ($Metric->"count" * 3)) / 2 = 0 && [ :typeof ($HostInfo->"pre-down-hook") ] = "str") do={
$NetwatchNotifyHook $HostName "pre-down" ($HostInfo->"pre-down-hook");
}
:if ($ParentNotified = false && $Metric->"count" >= $Count && $Metric->"notified" != true) do={
:local Message ("Host " . $HostName . " (" . $HostVal->"host" . ") is down since " . $HostVal->"since" . ".");
:if ([ :typeof ($HostInfo->"down-hook") ] = "str") do={
:set Message ($Message . "\n\n" . [ $NetwatchNotifyHook $HostName "down" ($HostInfo->"down-hook") ]);
}
:if ($HostInfo->"no-down-notification" != true) do={
$SendNotification2 ({ origin=$0; \
subject=([ $SymbolForNotification "cross-mark" ] . "Netwatch Notify: " . $HostName . " down"); \
message=$Message });
}
:set ($Metric->"notified") true;
}
}
:set ($NetwatchNotify->$HostName) {
"count"=($Metric->"count");
"notified"=($Metric->"notified");
"parent"=($Metric->"parent");
"resolve-failed"=($Metric->"resolve-failed");
"since"=($Metric->"since") };
}
}