Add postfix, re-create a new nagios server and further fix other issues

roles/nagios_config/tasks/main.yml

@@ -10,6 +10,13 @@
   - 'contacts'
   sudo: true
 
+- name: remove unwanted configure files
+  file: path={{ item }}_nagios2.cfg.j2 state=absent
+  with_items:
+  - 'localhost'
+  - 'extinfo'
+  sudo: true
+
 - name: change cgi config 
   template: src=cgi.cfg.j2 dest=/etc/nagios3/cgi.cfg
   sudo: true

roles/nagios_config/templates/hostgroups_nagios2.cfg.j2

 # Some generic hostgroup definitions
 
 {% for group in groups %}
+{% if group != "ungrouped" %}
 {% set nodelist = [] %}
 {% for node in groups[group] %}
 {% if nodelist.append(node) %}
@@ -10,4 +11,5 @@ define hostgroup {
     hostgroup_name {{ group }}
     members {{ nodelist|unique|join(',') }}
 }
+{% endif %}
 {% endfor %}

roles/nagios_monitored/files/scripts/check_apache2

+#!/usr/bin/python
+import sys, os, pwd
+import getopt
+import commands
+import subprocess
+
+STATE_OK=0
+STATE_WARNING=1
+
+check_apache=subprocess.Popen("service apache2 status", shell=True, stdout=subprocess.PIPE)
+apache_status=check_apache.communicate()[0]
+
+if "run" in apache_status:
+   print "Apache is Running"
+   sys.exit(STATE_OK)
+else:
+   print "Apache is NOT Running !!"   
+   sys.exit(STATE_WARNING)
+
+sys.exit(STATE_OK)
+
+
+

roles/nagios_monitored/files/scripts/check_disk

@@ -12,9 +12,14 @@ function check_disk() {
     local space="$1"
 
     local used=$(df -h ${space} | grep "^/" | awk '{print $5}' | sed -e 's/%//')
-    local threshhold=$((90))
+    local critical_threshhold=$((90))
+    local warning_threshhold=$((80))
 
-    if (( ${used} > ${threshhold} )); then
+    if (( ${used} > ${critical_threshhold} )); then
+        echo "CRITICAL: ${HOST_NAME} is running out of filesystem space"
+        exit "${STATE_CRITICAL}"
+    fi
+    if (( ${used} > ${warning_threshhold} )); then
         echo "WARNING: ${HOST_NAME} is running out of filesystem space"
         exit "${STATE_WARNING}"
     fi

roles/nagios_monitored/files/scripts/check_ldap

+#!/usr/bin/python
+import sys, os, pwd
+import getopt
+import commands
+import subprocess
+
+STATE_OK=0
+STATE_WARNING=1
+
+check_ldap=subprocess.Popen("service slapd status", shell=True, stdout=subprocess.PIPE)
+ldap_status=check_ldap.communicate()[0]
+
+if "run" in ldap_status:
+   print "Ldap is Running"
+   sys.exit(STATE_OK)
+else:
+   print "Ldap is NOT Running !!"   
+   sys.exit(STATE_WARNING)
+
+sys.exit(STATE_OK)
+
+
+

roles/nagios_monitored/files/scripts/check_ldap_client

+#!/bin/bash
+
+STATE_OK=0
+STATE_WARNING=1
+STATE_CRITICAL=2
+STATE_UNKNOWN=3
+STATE_DEPENDENT=4
+
+
+function main () {
+    id chines
+    local r=$?
+    if [ ${r} -ne 0 ]; then 
+        echo "Broken LDAP authentication"
+        return $STATE_CRITICAL
+    fi
+
+    echo "LDAP authentication OK"
+    return $STATE_OK
+
+}
+
+main
+exit $?

roles/nagios_monitored/files/scripts/check_mysql

+#!/usr/bin/python
+import sys, os, pwd
+import getopt
+import commands
+import subprocess
+
+STATE_OK=0
+STATE_WARNING=1
+
+check_mysql=subprocess.Popen("service mysqld status", shell=True, stdout=subprocess.PIPE)
+mysql_status=check_mysql.communicate()[0]
+
+if "run" in mysql_status:
+   print "Mysql is Running"
+   sys.exit(STATE_OK)
+else:
+   print "Mysql is NOT Running !!"   
+   sys.exit(STATE_WARNING)
+
+sys.exit(STATE_OK)
+
+
+

roles/nagios_monitored/files/scripts/check_slurm

+#!/usr/bin/python
+import sys, os, pwd
+import getopt
+import commands
+import subprocess
+
+STATE_OK=0
+STATE_WARNING=1
+
+check_slurm=subprocess.Popen("service slurm status", shell=True, stdout=subprocess.PIPE)
+slurm_status=check_slurm.communicate()[0]
+
+if "run" in slurm_status:
+   print "Slurm is Running"
+   sys.exit(STATE_OK)
+else:
+   print "Slurm is NOT Running !!"   
+   sys.exit(STATE_WARNING)
+
+sys.exit(STATE_OK)
+
+
+

roles/nagios_monitored/files/scripts/check_slurmdbd

+#!/usr/bin/python
+import sys, os, pwd
+import getopt
+import commands
+import subprocess
+
+STATE_OK=0
+STATE_WARNING=1
+
+check_slurmdbd=subprocess.Popen("service slurmdbd status", shell=True, stdout=subprocess.PIPE)
+slurmdbd_status=check_slurmdbd.communicate()[0]
+
+if "run" in slurmdbd_status:
+   print "Slurmdbd is Running"
+   sys.exit(STATE_OK)
+else:
+   print "Slurmdbd is NOT Running !!"   
+   sys.exit(STATE_WARNING)
+
+sys.exit(STATE_OK)
+

roles/nagios_monitored/tasks/main.yml

@@ -3,12 +3,16 @@
   user: name=nagios system=yes createhome=yes home={{ nagios_home }} shell=/bin/bash
   sudo: true
 
+- name: create ssh directory
+  file: path={{ nagios_home }}/.ssh state=directory owner=nagios mode=700
+  sudo: true
+
 - name: authorize_key
-  authorized_key: user=nagios key="{{ lookup('file', 'files/nagios_public_key') }}" path="{{ nagios_home }}"/.ssh/authorized_keys
+  authorized_key: user=nagios key="{{ lookup('file', 'files/nagios_public_key') }}" path="{{ nagios_home }}/.ssh/authorized_keys"
   sudo: true
 
 - name: make scripts directory
-  file: path=/var/lib/nagios/scripts state=directory owner=nagios mode=755
+  file: path={{ nagios_home }}/scripts state=directory owner=nagios mode=755
   sudo: true
 
 - name: install monitor scripts