X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=nagios-master.cfg;h=4560c013ad2258bf88c3acce833ece0181024fe6;hb=67462d48e72c5f49f4181d5fb1719ecf8e140107;hp=4e6975af07fd8e76dcff7fbcc2cb8e9856da3c81;hpb=b738fdef569643251a00f9e7d33831cce0a6c3fc;p=mirror%2Fdsa-nagios.git

diff --git a/nagios-master.cfg b/nagios-master.cfg
index 4e6975a..4560c01 100644
--- a/nagios-master.cfg
+++ b/nagios-master.cfg
@@ -15,11 +15,7 @@
 #  - verdi: pg upgrade, openvpn
 #  - mundy: salinfo_decode
 #  - puccini: mailgraph
-#  -
-#  - piatti
-#  - tartini
-#sarge:
-#  - spontini
+#  - lebrun: ippl
 
 # down:
 #  - ravel
@@ -37,6 +33,7 @@ servers:
     address: 82.195.75.126
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+    contacts: joerg, bzed
   gw-HP-ftc:
     address: 192.25.206.1
     parents: samosa
@@ -77,6 +74,7 @@ servers:
     address: 195.49.152.190
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+    contacts: bzed
   gw-freenet:
     address: 62.104.23.249
     parents: gw-HP-ftc
@@ -89,6 +87,7 @@ servers:
     address: 193.62.202.18
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+    contacts: tjrc1
   gw-cst:
     address: 213.188.99.215
     parents: gw-HP-ftc
@@ -102,15 +101,16 @@ servers:
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
   gw-1und1:
-    address: 195.20.247.53
+    address: 195.20.247.54
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+    contacts: joerg
   gw-blackcat:
     address: 193.201.200.129
     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:
@@ -129,6 +129,24 @@ servers:
     address: 65.173.90.18
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+  gw-ubc:
+    address: 137.82.84.41
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+    contacts: lfilipoz
+  gw-carnet:
+    address: 161.53.160.1
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-telegraaf:
+    address: 82.94.249.153
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-helsinki:
+    address: 128.214.173.25
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+    contacts: holger
 
   samosa:
     address: 192.25.206.57
@@ -153,10 +171,12 @@ servers:
     address: 192.25.206.15
     parents: samosa
     hostgroups: computers, buildd, sw-raid, single-cpu
+    contacts: dannf
   penalosa:
     address: 192.25.206.68
     parents: samosa
     hostgroups: computers, buildd, sw-raid, single-cpu
+    contacts: dannf
   mundy:
     address: 192.25.206.62
     parents: samosa
@@ -174,22 +194,27 @@ servers:
     address: 82.195.75.91
     parents: gw-man-da
     hostgroups: computers, service, syslog-ng-hosts, postfix-hosts, dl385
+    contacts: joerg, bzed
   sperger:
     address: 82.195.75.98
     parents: gw-man-da
     hostgroups: computers, porterbox, sw-raid
+    contacts: bzed
   agricola:
     address: 82.195.75.86
     parents: gw-man-da
     hostgroups: computers, porterbox, sw-raid, single-cpu
+    contacts: bzed
   arcadelt:
     address: 82.195.75.87
     parents: gw-man-da
     hostgroups: computers, buildd, sw-raid, single-cpu
+    contacts: bzed
   liszt:
     address: 82.195.75.100
     parents: gw-man-da
     hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, dl385
+    contacts: bzed
 
   master:
     address: 70.103.162.29
@@ -198,7 +223,7 @@ 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
@@ -258,6 +283,7 @@ servers:
     address: 195.49.152.174
     parents: gw-dg-i.net
     hostgroups: computers, buildd, sw-raid, single-cpu
+    contacts: bzed
 
   pergolesi:
     address: 62.104.23.252
@@ -271,16 +297,18 @@ servers:
   raptor:
     address: 195.243.109.162
     parents: gw-topalis
-    hostgroups: computers, porterbox
+    hostgroups: computers, porterbox, postfix-hosts
 
   albeniz:
     address: 193.62.202.27
     parents: gw-sanger
     hostgroups: computers, porterbox, sw-raid
+    contacts: tjrc1
   goetz:
     address: 193.62.202.26
     parents: gw-sanger
     hostgroups: computers, buildd, sw-raid
+    contacts: tjrc1
 
   escher:
     address: 213.188.99.215
@@ -296,23 +324,25 @@ servers:
     address: 72.66.115.54
     parents: gw-frost
     hostgroups: computers, buildd
+    contacts: sfrost
 
   puccini:
     address: 87.106.4.56
     parents: gw-1und1
     hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, amavis-hosts
+    contacts: joerg
 
   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
 
@@ -320,10 +350,12 @@ servers:
     address: 217.114.76.82
     parents: gw-nmmn
     hostgroups: deadslow
+    contacts: luk
   crest:
     address: 217.114.76.83
     parents: gw-nmmn
     hostgroups: deadslow
+    contacts: luk
 
   kassia:
     address: 130.89.175.54
@@ -334,12 +366,35 @@ servers:
     address: 157.193.39.233
     parents: gw-HP-ftc
     hostgroups: computers, buildd, postfix-hosts, sw-raid, single-cpu
+    contacts: luk
 
   agnesi:
     address: 65.173.90.83
     parents: gw-agnesi
     hostgroups: deadslow
 
+  spontini:
+    address: 137.82.84.42
+    parents: gw-ubc
+    hostgroups: computers, buildd
+    contacts: lfilipoz
+
+  lebrun:
+    address: 161.53.160.165
+    parents: gw-carnet
+    hostgroups: computers, buildd
+
+  tartini:
+    address: 82.94.249.158
+    parents: gw-telegraaf
+    hostgroups: computers, sw-raid
+
+  piatti:
+    address: 193.167.161.225
+    parents: gw-helsinki
+    hostgroups: computers, postfix-hosts, dl385
+    contacts: holger
+
 #############################
 # host groups
 #
@@ -439,6 +494,20 @@ hostgroups:
     alias: secondary IP addresses
     private: 1
 
+
+#############################
+# servicegroups
+#############################
+servicegroups:
+  diskspace:
+    alias: diskusage checks
+  buildd:
+    alias: buildd checks
+  raid:
+    alias: raid checks
+  kernel:
+    alias: kernel checks
+
 #############################
 # services
 #############################
@@ -466,62 +535,82 @@ services:
  ####
   -
     name: disk usage - all
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 90 95"
     hostgroups: computers
   -
     name: disk usage on /
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /"
     hostgroups: computers
   -
     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
+    hosts: sperger, rietz, steffani, penalosa, peri, albeniz, escher, goetz, mayer, mayr, paer, spontini, tartini
   -
     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
+    hosts: bartok, samosa, raff, lobos, villa, gluck, saens, escher, voltaire, puccini, lebrun, tartini
   -
     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
+    hosts: bartok, sperger, samosa, raff, lobos, villa, steffani, saens, pergolesi, verdi, puccini, spontini
   -
     name: disk usage on /org
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /org"
     hosts: merkel
   -
     name: disk usage on /srv
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /srv"
-    hosts: agricola, arcadelt, argento, allegri
+    hosts: agricola, arcadelt, argento, allegri, tartini
   -
     name: disk usage on /org/scratch
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org/scratch"
     hosts: merkel
   -
     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
+    hosts: samosa, raff, gluck, saens, escher, 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
+    hosts: samosa, raff, lobos, villa, gluck, saens, pergolesi, puccini, merulo, tartini
   -
     name: disk usage on /home
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /home"
-    hosts: raptor, escher, voltaire
+    hosts: raptor, escher, voltaire, lebrun
   -
     name: disk usage on /home
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /home"
     hosts: gluck
   -
     name: disk usage on /chroot
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /chroot"
     hosts: raptor
   -
     name: disk usage on /mnt/hdc
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /mnt/hdc"
     hosts: voltaire
+  -
+    name: disk usage on /mnt/sdb1
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /mnt/sdb1"
+    hosts: spontini
   -
     name: disk usage on /x
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /x"
     hosts: caballero
 
@@ -544,6 +633,15 @@ 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
@@ -600,6 +698,7 @@ services:
     hostgroups: computers
     depends: process - sshd
     normal_check_interval:  60
+    notification_interval: 1440
 
   -
     name: "network service - sshd"
@@ -612,16 +711,23 @@ 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 - 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
  ###
   -
@@ -662,7 +768,7 @@ services:
  ###
   -
     name: process - cron
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C cron -a /usr/sbin/cron"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u root -C cron -a /usr/sbin/cron"
     hostgroups: computers
 
  ###
@@ -848,7 +954,7 @@ services:
     depends: process - postfix - master
   -
     name: process - postfix - anvil
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C anvil -a 'anvil -l -t unix -u'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:1 -c 0: -u postfix -C anvil -a 'anvil -l -t unix -u'"
     hostgroups: postfix-hosts
     depends: process - postfix - master
 
@@ -875,7 +981,7 @@ services:
 
   -
     name: process - postfix - smtpd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:20 -c 0:50 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:40 -c 0:90 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'"
     hostgroups: postfix-hosts
     excludehosts: liszt
     depends: process - postfix - master
@@ -909,12 +1015,23 @@ services:
     name: network service - smtp
     check: dsa_check_smtp
     hostgroups: postfix-hosts
-    excludehosts: verdi, kassia, allegri
+    excludehosts: verdi, kassia, allegri, raptor, piatti
     depends: process - postfix - master
   -
     name: network service - smtp - port 2025
     check: dsa_check_smtp_port!2025
-    hosts: verdi, kassia, murphy, allegri
+    hosts: verdi, kassia, murphy, allegri, piatti
+    depends: process - postfix - master
+  -
+    name: network service - smtp - port 8080
+    check: dsa_check_smtp_port!8080
+    hosts: murphy, piatti
+    depends: process - postfix - master
+  -
+    name: network service - smtp - port 2025
+    remotecheck: /usr/lib/nagios/plugins/check_smtp -t 40 -H $HOSTADDRESS$ -p 2025
+    runfrom: murphy
+    hosts: raptor
     depends: process - postfix - master
 
   -
@@ -922,14 +1039,18 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_http -H localhost -p 465 -S -C 14 -t 45"
     hostgroups: postfix-hosts
     depends: process - postfix - master
-    normal_check_interval: 1440
+    normal_check_interval: 120
 
 
   -
     name: setup - debian-admin in etc aliases
     nrpe: "/usr/lib/nagios/plugins/dsa-check-da-in-aliases"
     hostgroups: computers
-    normal_check_interval: 1440
+    normal_check_interval: 120
+  -
+    name: setup - ud-ldap freshness
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-udldap-freshness"
+    hostgroups: computers
  ###
   -
     name: process - uptimed
@@ -942,7 +1063,7 @@ services:
     hostgroups: computers
     excludehostgroups: single-cpu
   -
-    name: unwanted process - named
+    name: unwanted process - irqbalance
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance"
     hostgroups: single-cpu
 
@@ -995,22 +1116,49 @@ services:
  ###
   -
     name: process - mdadm monitor
+    servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C mdadm -a '/sbin/mdadm --monitor --pid-file /var/run/mdadm/monitor.pid --daemonise --scan'"
     hostgroups: sw-raid
   -
     name: RAID - sw raid
+    servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-sw"
     hostgroups: sw-raid
 
  ###
   -
     name: process - cpqarrayd
+    servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C cpqarrayd -a '/usr/sbin/cpqarrayd'"
     hostgroups: dl385, dl380, dl360
   -
     name: RAID - arrayprobe
+    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"
+    hostgroups: dl385, dl380, dl360
+ ###
+  -
+    name: RAID - DAC960
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-dac960"
+    hosts: verdi
+ ###
+  -
+    name: RAID - 3ware
+    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
 
  ###
   -
@@ -1060,10 +1208,6 @@ services:
     depends: rietz:process - xinetd
 
  ###
-  -
-    name: process - nagios1
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nagios -C nagios -a '/usr/sbin/nagios -d /etc/nagios/nagios.cfg'"
-    hosts: samosa
   -
     name: process - nagios3
     # there is always one extra process per check currently running..
@@ -1155,7 +1299,7 @@ services:
  ####
   -
     name: process - debianqueued
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u dak -C debianqueued -a '/usr/bin/perl -w ./debianqueued'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u dak -C debianqueued"
     hosts: ries
 
  ###
@@ -1180,7 +1324,21 @@ services:
     depends: process - postresql81 - master
 
  ####
-  # XXX is this needed
+  -
+    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?
   -
     name: process - snmpd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u snmp -C snmpd -a '/usr/sbin/snmpd -Lsd -Lf /dev/null -u snmp -I -smux -p /var/run/snmpd.pid 127.0.0.1'"
@@ -1189,6 +1347,7 @@ services:
  ###
   -
     name: process - buildd
+    servicegroups: buildd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u buildd -C buildd '/usr/bin/perl /usr/bin/buildd'"
     hostgroups: buildd