From 01bb987fb328d46c7d7f77934aaebd9d8c46382f Mon Sep 17 00:00:00 2001
From: Victor LABORIE <vlaborie@evolix.fr>
Date: Wed, 17 May 2017 10:32:39 +0200
Subject: [PATCH] nagios-nrpe: add varnish check and tags

---
 .../files/plugins/check_varnish_health        | 89 +++++++++++++++++++
 nagios-nrpe/tasks/main.yml                    |  4 +
 nagios-nrpe/templates/evolix.cfg.j2           |  3 +-
 3 files changed, 95 insertions(+), 1 deletion(-)
 create mode 100755 nagios-nrpe/files/plugins/check_varnish_health

diff --git a/nagios-nrpe/files/plugins/check_varnish_health b/nagios-nrpe/files/plugins/check_varnish_health
new file mode 100755
index 00000000..6b2d883a
--- /dev/null
+++ b/nagios-nrpe/files/plugins/check_varnish_health
@@ -0,0 +1,89 @@
+#!/bin/bash
+#
+# check_varnish_health
+#
+# A nagios plugin to check the health status of a varnish web cache using varnishadm.
+# by rudi kramer
+# Copyright 2010 rudi Kramer <rudi(dot)kramer(at)gmail(dot)com>, BSD-style copyright and disclaimer.
+#
+# version 1.1 - Fixed bug where backend server is not called default 
+# 
+# Command Line Usage: ./check_varnish_health -i <ip> -p <port>
+#
+# Example: ./check_varnish_health -i 127.0.0.1 -p 6082
+#
+#################################################################################
+
+PATH=/sbin:/bin:/usr/sbin:/usr/bin
+
+E_SUCCESS="0"
+E_WARNING="1"
+E_CRITICAL="2"
+E_UNKNOWN="3"
+
+PROGNAME=`basename $0`
+CHECK_VARNISHD=`ps ax| grep varnishd| grep -v grep`
+
+print_help() {
+        echo ""
+        echo "This plugin checks the health status of a varnish web cache using varnishadm"
+        echo ""
+        echo "Usage: $PROGNAME -i <ip address>  -p <port> -s <secret file> -w <maximum number of sick backends before raise a warning state> -c <same as previous but for critical state>"
+        echo ""
+}
+
+if [ $# -lt 10 ]; then
+        print_help
+        exit $E_UNKNOWN
+fi
+
+if [ -z "$CHECK_VARNISHD" ]; then
+    echo "CRITICAL: varnishd is not running, unable to check health status"
+    exit $E_CRITICAL
+fi
+
+while test -n "$1"; do
+    case "$1" in
+        --help | -h)
+            print_help
+            exit $E_UNKNOWN
+            ;;
+        -i)
+            IP=$2
+            shift
+            ;;
+        -p)
+            PORT=$2
+            shift
+            ;;
+        -s)
+            SECRET=$2
+            shift
+            ;;
+        -w)
+            WARNING=$2
+            shift
+            ;;
+        -c)
+            CRITICAL=$2
+            shift
+            ;;
+    esac
+    shift
+done
+
+COMMAND=`varnishadm -S $SECRET -T $IP\:$PORT debug.health  2>&1 | grep "^Backend " | grep -v "Healthy" |wc -l`
+OUTPUT=`varnishadm -S $SECRET -T $IP\:$PORT debug.health  2>&1 | grep "^Backend " |tr '\n' ','`
+
+if [ $COMMAND -lt $WARNING ]; then
+    echo "OK: $OUTPUT"
+    exit ${E_SUCCESS}
+
+elif [ $COMMAND -lt $CRITICAL ]; then
+    echo "WARNING: $OUTPUT"
+    exit ${E_WARNING}
+
+else
+    echo "CRITICAL: $OUTPUT"
+    exit ${E_CRITICAL}
+fi
diff --git a/nagios-nrpe/tasks/main.yml b/nagios-nrpe/tasks/main.yml
index bb765176..e723d322 100644
--- a/nagios-nrpe/tasks/main.yml
+++ b/nagios-nrpe/tasks/main.yml
@@ -1,6 +1,10 @@
 ---
 - include: debian.yml
   when: ansible_os_family == "Debian"
+  tags:
+  - nagios
 
 - include: openbsd.yml
   when: ansible_os_family == "OpenBSD"
+  tags:
+  - nagios
diff --git a/nagios-nrpe/templates/evolix.cfg.j2 b/nagios-nrpe/templates/evolix.cfg.j2
index 24f569f1..96bc5649 100644
--- a/nagios-nrpe/templates/evolix.cfg.j2
+++ b/nagios-nrpe/templates/evolix.cfg.j2
@@ -45,6 +45,7 @@ command[check_redis]=/usr/lib/nagios/plugins/check_tcp -p 6379
 command[check_clamd]=/usr/lib/nagios/plugins/check_clamd -H /var/run/clamav/clamd.ctl -v
 command[check_ssl]=/usr/lib/nagios/plugins/check_http -f follow -I 127.0.0.1 -S -p 443 -H ssl.evolix.net -C 15,5
 command[check_elasticsearch]=/usr/lib/nagios/plugins/check_http -H localhost -u /_cluster/health -p 9200 -r '"status":"red",' --invert-regex
+command[check_memcached]=/usr/lib/nagios/plugins/check_tcp -H 127.0.0.1 -p 11211
 
 # Local checks (not packaged)
 command[check_mem]={{ nagios_plugins_directory }}/check_mem -f -C -w 20 -c 10
@@ -57,7 +58,7 @@ command[check_drbd]={{ nagios_plugins_directory }}/check_drbd -d All -c StandAlo
 command[check_mongodb_connect]={{ nagios_plugins_directory }}/check_mongodb -H localhost -P27017 -A connect
 command[check_glusterfs]={{ nagios_plugins_directory }}/check_glusterfs -v all -n 0
 command[check_supervisord_status]={{ nagios_plugins_directory }}/check_supervisord
-command[check_memcached]=/usr/lib/nagios/plugins/check_tcp -H 127.0.0.1 -p 11211
+command[check_varnish]={{ nagios_plugins_directory }}/check_varnish_health -i 127.0.0.1 -p 6082 -s /etc/varnish/secret -w 2 -c 4
 
 # Check HTTP "many". Use this to check many websites (http, https, ports, sockets and SSL certificates).
 # Beware! All checks must not take more than 10s!