X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=nagios-master.cfg;h=53a2c20593a994ad0a2646b6f1f227fbaca25bc7;hb=363f9dbade0e4ef56c4f443b9887373ef3de7ab2;hp=790335693a06a330528b30117b4306f9e11f9d26;hpb=5a2ba5b3e57061e6a6db59ce30a9e4ea337773ee;p=mirror%2Fdsa-nagios.git

diff --git a/nagios-master.cfg b/nagios-master.cfg
index 7903356..53a2c20 100644
--- a/nagios-master.cfg
+++ b/nagios-master.cfg
@@ -18,11 +18,9 @@
 #  - lebrun: ippl
 
 # down:
-#  - ravel
 #  - sarti
 #  - goedel
 #  - leisner
-#  - schulz
 
 ---
 #############################
@@ -88,10 +86,6 @@ servers:
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
     contacts: tjrc1
-  gw-cst:
-    address: 213.188.99.215
-    parents: gw-HP-ftc
-    hostgroups: routing-infrastructure
   gw-lrz:
     address: 129.187.0.150
     parents: gw-HP-ftc
@@ -110,7 +104,7 @@ servers:
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
   gw-xandros:
-    address: 142.46.212.33
+    address: 67.210.160.81
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
   gw-nmmn:
@@ -193,8 +187,13 @@ servers:
   bartok:
     address: 82.195.75.91
     parents: gw-man-da
-    hostgroups: computers, service, syslog-ng-hosts, postfix-hosts, dl385
+    hostgroups: computers, service, syslog-ng-hosts, postfix-hosts, dl385, bind9-hosts
     contacts: joerg, bzed
+  morales:
+    address: 82.195.75.97
+    parents: gw-man-da
+    hostgroups: computers, porterbox, single-cpu
+    contacts: bzed
   sperger:
     address: 82.195.75.98
     parents: gw-man-da
@@ -203,17 +202,22 @@ servers:
   agricola:
     address: 82.195.75.86
     parents: gw-man-da
-    hostgroups: computers, porterbox, sw-raid, single-cpu
+    hostgroups: computers, porterbox, sw-raid, single-cpu, lenny
     contacts: bzed
   arcadelt:
     address: 82.195.75.87
     parents: gw-man-da
-    hostgroups: computers, buildd, sw-raid, single-cpu
+    hostgroups: computers, buildd, sw-raid, single-cpu, lenny
     contacts: bzed
   liszt:
     address: 82.195.75.100
     parents: gw-man-da
-    hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, dl385
+    hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, amavis-hosts, dl385
+    contacts: bzed
+  auber:
+    address: 82.195.75.101
+    parents: gw-man-da
+    hostgroups: computers, buildd, sw-raid
     contacts: bzed
 
   master:
@@ -223,7 +227,7 @@ servers:
   murphy:
     address: 70.103.162.31
     parents: gw-brainfood
-    hostgroups: computers, general, apache2-hosts, bind9-hosts, postfix-hosts
+    hostgroups: computers, general, apache2-hosts, bind9-hosts, postfix-hosts, dl380
 
   ries:
     address: 128.148.34.103
@@ -263,11 +267,11 @@ servers:
   steffani:
     address: 128.31.0.36
     parents: gw-mit-csail
-    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, sw-raid
-  casals:
-    address: 128.31.0.16
+    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, sw-raid, bind9-hosts
+  mahler:
+    address: 128.31.0.46
     parents: gw-mit-csail
-    hostgroups: deadslow
+    hostgroups: computers, single-cpu, lenny, porterbox, rsyslog-hosts
 
   klecker:
     address: 194.109.137.218
@@ -282,7 +286,7 @@ servers:
   argento:
     address: 195.49.152.174
     parents: gw-dg-i.net
-    hostgroups: computers, buildd, sw-raid, single-cpu
+    hostgroups: computers, buildd, sw-raid, single-cpu, lenny
     contacts: bzed
 
   pergolesi:
@@ -302,19 +306,16 @@ servers:
   albeniz:
     address: 193.62.202.27
     parents: gw-sanger
-    hostgroups: computers, porterbox, sw-raid
+    # SMP kernel doesn't run stable
+    hostgroups: computers, porterbox, sw-raid, single-cpu
     contacts: tjrc1
   goetz:
     address: 193.62.202.26
     parents: gw-sanger
-    hostgroups: computers, buildd, sw-raid
+    # SMP kernel doesn't run stable
+    hostgroups: computers, buildd, sw-raid, single-cpu
     contacts: tjrc1
 
-  escher:
-    address: 213.188.99.215
-    parents: gw-cst
-    hostgroups: computers, porterbox, single-cpu
-
   verdi:
     address: 192.54.42.193
     parents: gw-lrz
@@ -335,14 +336,14 @@ servers:
   caballero:
     address: 193.201.200.200
     parents: gw-blackcat
-    hostgroups: computers, buildd, sw-raid
+    hostgroups: computers, buildd, sw-raid, bind9-hosts
 
   elara:
-    address: 142.46.212.46
+    address: 67.210.160.90
     parents: gw-xandros
     hostgroups: deadslow
   europa:
-    address: 142.46.212.46
+    address: 67.210.160.89
     parents: gw-xandros
     hostgroups: deadslow
 
@@ -365,7 +366,7 @@ servers:
   allegri:
     address: 157.193.39.233
     parents: gw-HP-ftc
-    hostgroups: computers, buildd, postfix-hosts, sw-raid, single-cpu
+    hostgroups: computers, buildd, postfix-hosts, sw-raid, single-cpu, lenny
     contacts: luk
 
   agnesi:
@@ -378,6 +379,10 @@ servers:
     parents: gw-ubc
     hostgroups: computers, buildd
     contacts: lfilipoz
+  ravel:
+    address: 137.82.84.43
+    parents: gw-ubc
+    hostgroups: computers, general, dl385, apache2-hosts
 
   lebrun:
     address: 161.53.160.165
@@ -387,7 +392,7 @@ servers:
   tartini:
     address: 82.94.249.158
     parents: gw-telegraaf
-    hostgroups: computers
+    hostgroups: computers, sw-raid, apache2-hosts, mysql-hosts
 
   piatti:
     address: 193.167.161.225
@@ -449,9 +454,16 @@ hostgroups:
     alias: Hosts with only one CPU
     private: 1
 
+  lenny:
+    alias: Hosts running lenny, not etch
+    private: 1
+
   syslog-ng-hosts:
     alias: hosts running syslog-ng instead of sysklogd
     private: 1
+  rsyslog-hosts:
+    alias: hosts running rsyslogd instead of sysklogd
+    private: 1
   postfix-hosts:
     alias: hosts running postfix instead of exim
     private: 1
@@ -459,7 +471,7 @@ hostgroups:
     alias: "hosts running the full mail stuff, including clamav, SA, and greylistd"
     private: 1
   heavy-postfix:
-    alias: "postfix hosts running the full mail stuff, including clamav, SA, postgrey, amavis, policyd-weight"
+    alias: "postfix hosts running the full mail stuff, including clamav, SA, postgrey, policyd-weight"
     private: 1
   apache2-hosts:
     alias: hosts running apache2
@@ -485,6 +497,9 @@ hostgroups:
   postgres81-hosts:
     alias: hosts running postgres81
     private: 1
+  mysql-hosts:
+    alias: hosts running mysql
+    private: 1
 
   highload:
     alias: "hosts on which high load is normal"
@@ -505,6 +520,15 @@ servicegroups:
     alias: buildd checks
   raid:
     alias: raid checks
+  kernel:
+    alias: kernel checks
+  weaksshkeys:
+    alias: weak ssh keys
+  apt:
+    alias: apt upgrade status
+  security:
+    alias: security
+    servicegroup_members: apt, weaksshkeys, kernel
 
 #############################
 # services
@@ -541,31 +565,37 @@ services:
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /"
     hostgroups: computers
+    excludehosts: ries
+  -
+    name: disk usage on /
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /"
+    hosts: ries
   -
     name: disk usage on /boot
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /boot"
-    hosts: sperger, rietz, steffani, penalosa, peri, albeniz, escher, goetz, mayer, mayr, paer, spontini, tartini
+    hosts: sperger, rietz, steffani, penalosa, peri, albeniz, goetz, mayer, mayr, paer, spontini, tartini, morales, ravel, auber
   -
     name: disk usage on /var
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /var"
-    hosts: bartok, samosa, raff, lobos, villa, gluck, saens, escher, voltaire, puccini, lebrun, tartini
+    hosts: bartok, samosa, raff, lobos, villa, gluck, saens, voltaire, puccini, lebrun, tartini, morales
   -
     name: disk usage on /org
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org"
-    hosts: bartok, sperger, samosa, raff, lobos, villa, steffani, saens, pergolesi, verdi, puccini, spontini
+    hosts: sperger, samosa, raff, lobos, villa, steffani, saens, pergolesi, verdi, puccini, spontini, ravel, mahler
   -
     name: disk usage on /org
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /org"
-    hosts: merkel
+    hosts: merkel, bartok
   -
     name: disk usage on /srv
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /srv"
-    hosts: agricola, arcadelt, argento, allegri, tartini
+    hosts: agricola, arcadelt, argento, allegri, tartini, morales, auber
   -
     name: disk usage on /org/scratch
     servicegroups: diskspace
@@ -575,17 +605,17 @@ services:
     name: disk usage on /tmp
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /tmp"
-    hosts: samosa, raff, gluck, saens, escher, puccini, merkel, tartini
+    hosts: samosa, raff, gluck, saens, puccini, merkel, tartini
   -
     name: disk usage on /usr
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /usr"
-    hosts: samosa, raff, lobos, villa, gluck, saens, pergolesi, puccini, merulo, tartini
+    hosts: samosa, raff, lobos, villa, gluck, saens, pergolesi, puccini, merulo, tartini, morales
   -
     name: disk usage on /home
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /home"
-    hosts: raptor, escher, voltaire, lebrun
+    hosts: raptor, voltaire, lebrun
   -
     name: disk usage on /home
     servicegroups: diskspace
@@ -614,14 +644,13 @@ services:
 
  ############ All Computers ############
  ####
-  #-
-  #  name: apt - security updates
-  #  nrpe: "/usr/local/bin/nagios-check-apt-updates --warnifupdates"
-  #  hostgroups: computers
-  #  normal_check_interval: 480
-  #  notification_interval: 480
-  #  max_check_attempts: 4
-  #  retry_check_interval: 12
+  -
+    name: apt - security updates
+    servicegroups: apt
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-statusfile /var/cache/dsa/nagios/apt"
+    hostgroups: computers
+    normal_check_interval:  60
+    retry_check_interval: 15
  ####
   -
     name: backup
@@ -631,6 +660,15 @@ services:
     max_check_attempts: 2
     retry_check_interval: 5
 
+ ####
+  -
+    name: running kernel
+    servicegroups: kernel
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-running-kernel"
+    hostgroups: computers
+    normal_check_interval: 180
+    retry_check_interval: 5
+
  ####
   -
     name: users
@@ -687,6 +725,7 @@ services:
     hostgroups: computers
     depends: process - sshd
     normal_check_interval:  60
+    notification_interval: 1440
 
   -
     name: "network service - sshd"
@@ -699,16 +738,57 @@ services:
     check: dsa_check_ssh_port!2260
     hosts: agnesi
     normal_check_interval:  180
+  -
+    name: "network service - sshd - 443"
+    check: dsa_check_ssh_port!443
+    hosts: gluck
+    normal_check_interval:  180
+
+  -
+    name: "network service - sshd - version"
+    check: "dsa_check_ssh_port_version!22!OpenSSH_4.3p2 Debian-9etch2"
+    depends: network service - sshd
+    hostgroups: computers, deadslow
+    excludehosts: agnesi, crest, kullervo
+    excludehostgroups: lenny
+    normal_check_interval:  360
+  -
+    name: "network service - sshd - version"
+    check: "dsa_check_ssh_port_version!22!OpenSSH_4.3p2 Debian-9etch2+m68k1"
+    depends: network service - sshd
+    hosts: crest, kullervo
+    normal_check_interval:  360
+  -
+    name: "network service - sshd - version"
+    check: "dsa_check_ssh_port_version!22!OpenSSH_5.1p1 Debian-2"
+    depends: network service - sshd
+    hostgroups: lenny
+    normal_check_interval:  360
+  -
+    name: "network service - sshd - version - 2260"
+    check: "dsa_check_ssh_port_version!2260!OpenSSH_4.3p2 Debian-9etch2"
+    depends: network service - sshd - 2260
+    hosts: agnesi
+    normal_check_interval:  360
+ #
+  -
+    name: ssh - weak keys
+    servicegroups: weaksshkeys
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-statusfile /var/cache/dsa/nagios/weak-ssh-keys"
+    hostgroups: computers
+    normal_check_interval:  360
  ####
   -
     name: network service - nrpe
     check: check_tcp!5666
     hostgroups: computers
-    max_check_attempts: -1
+    max_check_attempts: -2
+    notification_interval: 1440
   -
     name: process - nrpe
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u nagios -C nrpe -a '/usr/sbin/nrpe -c /etc/nagios/nrpe.cfg -d'"
     hostgroups: computers
+    max_check_attempts: -1
     depends: network service - nrpe
  ###
   -
@@ -757,17 +837,22 @@ services:
     name: process - syslog-ng
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C syslog-ng  -a '/sbin/syslog-ng -p /var/run/syslog-ng.pid'"
     hostgroups: syslog-ng-hosts
+ ###
+  -
+    name: process - rsyslogd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C rsyslogd  -a '/usr/sbin/rsyslogd -c3'"
+    hostgroups: rsyslog-hosts
  ###
   -
     name: process - syslogd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C syslogd  -a '/sbin/syslogd'"
     hostgroups: computers
-    excludehostgroups: syslog-ng-hosts
+    excludehostgroups: syslog-ng-hosts, rsyslog-hosts
   -
     name: process - klogd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C klogd  -a '/sbin/klogd -x'"
     hostgroups: computers
-    excludehostgroups: syslog-ng-hosts
+    excludehostgroups: syslog-ng-hosts, rsyslog-hosts
 
  ### MAIL STUFF
  ###
@@ -801,6 +886,10 @@ services:
     name: process - clamav - freshclam
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u clamav -C freshclam -a '/usr/bin/freshclam -d --quiet'"
     hostgroups: heavy-exim, heavy-postfix
+  -
+    name: process - clamav - getsigs
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u clamav -C getclamsigs -a 'getclamsigs'"
+    hostgroups: heavy-exim, heavy-postfix
   #
   -
     name: unwanted process - clamav
@@ -812,6 +901,11 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C freshclam"
     hostgroups: computers
     excludehostgroups: heavy-exim, heavy-postfix
+  -
+    name: unwanted process - clamav - getsigs
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C getclamsigs"
+    hostgroups: computers
+    excludehostgroups: heavy-exim, heavy-postfix
  ###
   -
     name: process - spamd - master
@@ -1087,7 +1181,7 @@ services:
   -
     name: process - rngd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C rngd  -a '/usr/sbin/rngd -r /dev/hwrng'"
-    hosts: bartok
+    hostgroups: dl385
  ###
   -
     name: process - sensord
@@ -1117,6 +1211,12 @@ services:
     servicegroups: raid
     nrpe: "sudo /usr/bin/arrayprobe"
     hostgroups: dl385, dl380, dl360
+  -
+    name: HW - hpacucli status
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-hpacucli"
+    normal_check_interval: 120
+    hostgroups: dl385, dl380, dl360
  ###
   -
     name: RAID - DAC960
@@ -1129,17 +1229,23 @@ services:
     servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-3ware"
     hosts: puccini
+ ###
+  -
+    name: RAID - MPT
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-mpt"
+    hosts: master
 
  ###
   -
     name: process - slapd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:20 -c 1:50 -u openldap -C slapd -a '/usr/sbin/slapd -g openldap -u openldap'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:20 -c 1:50 -u openldap -C slapd -a '/usr/sbin/slapd -h ldap:/// ldaps:/// -g openldap -u openldap'"
     hosts: samosa
  ###
   -
     name: process - udevd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C udevd -a 'udevd'"
-    hosts: sperger, ries, steffani, merkel, spohr, peri, penalosa, albeniz, escher, verdi, liszt, kassia, agricola, arcadelt, argento, allegri
+    hosts: sperger, ries, steffani, merkel, spohr, peri, penalosa, albeniz, verdi, liszt, kassia, agricola, arcadelt, argento, allegri, morales, bartok
  ###
   -
     name: process - acpid
@@ -1292,21 +1398,31 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postmaster -a 'postgres: stats collector process'"
     hostgroups: postgres81-hosts
     depends: process - postresql81 - master
-
  ####
   -
-    name: process - xenconsoled
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xenconsoled -a 'xenconsoled'"
-    hosts: piatti
+    name: process - mysql - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u root -C mysqld_safe -a '/bin/sh /usr/bin/mysqld_safe'"
+    hostgroups: mysql-hosts
   -
-    name: process - xenstored
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xenstored -a '/usr/lib/xen-3.0.3-1/bin/xenstored --pid-file /var/run/xenstore.pid'"
-    hosts: piatti
-  -
-    name: process - xend
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 2: -u root -C python -a 'python /usr/lib/xen-3.0.3-1/bin/xend start'"
-    hosts: piatti
+    name: process - mysql - workers
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u mysql -C mysqld -a /usr/sbin/mysqld"
+    hostgroups: mysql-hosts
+    depends: process - mysql - master
 
+ ####
+  #-
+  #  name: process - xenconsoled
+  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xenconsoled -a 'xenconsoled'"
+  #  hosts: piatti
+  #-
+  #  name: process - xenstored
+  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xenstored -a '/usr/lib/xen-3.0.3-1/bin/xenstored --pid-file /var/run/xenstore.pid'"
+  #  hosts: piatti
+  #-
+  #  name: process - xend
+  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 2: -u root -C python -a 'python /usr/lib/xen-3.0.3-1/bin/xend start'"
+  #  hosts: piatti
+#
  ####
   # XXX is this needed?
   -