X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=nagios-master.cfg;h=4560c013ad2258bf88c3acce833ece0181024fe6;hb=67462d48e72c5f49f4181d5fb1719ecf8e140107;hp=26030e1a27de2003f21d30280e528f644654abb7;hpb=308d8c4ec3121ce856ba8a187c86c62de97ebbcb;p=mirror%2Fdsa-nagios.git diff --git a/nagios-master.cfg b/nagios-master.cfg index 26030e1..4560c01 100644 --- a/nagios-master.cfg +++ b/nagios-master.cfg @@ -16,9 +16,6 @@ # - mundy: salinfo_decode # - puccini: mailgraph # - lebrun: ippl -# - -# - piatti -# - tartini # down: # - ravel @@ -113,7 +110,7 @@ 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: @@ -136,10 +133,20 @@ servers: 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 @@ -216,7 +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 @@ -290,7 +297,7 @@ servers: raptor: address: 195.243.109.162 parents: gw-topalis - hostgroups: computers, porterbox + hostgroups: computers, porterbox, postfix-hosts albeniz: address: 193.62.202.27 @@ -317,6 +324,7 @@ servers: address: 72.66.115.54 parents: gw-frost hostgroups: computers, buildd + contacts: sfrost puccini: address: 87.106.4.56 @@ -327,14 +335,14 @@ 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 @@ -342,10 +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 @@ -356,6 +366,7 @@ 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 @@ -366,12 +377,24 @@ servers: 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 # @@ -480,6 +503,10 @@ servicegroups: alias: diskusage checks buildd: alias: buildd checks + raid: + alias: raid checks + kernel: + alias: kernel checks ############################# # services @@ -520,12 +547,12 @@ services: 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 + 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, lebrun + hosts: bartok, samosa, raff, lobos, villa, gluck, saens, escher, voltaire, puccini, lebrun, tartini - name: disk usage on /org servicegroups: diskspace @@ -540,7 +567,7 @@ services: 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 @@ -550,12 +577,12 @@ 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 + 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 @@ -606,6 +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 @@ -662,6 +698,7 @@ services: hostgroups: computers depends: process - sshd normal_check_interval: 60 + notification_interval: 1440 - name: "network service - sshd" @@ -674,16 +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 ### - @@ -910,7 +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 @@ -971,12 +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 - @@ -1008,7 +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 @@ -1061,22 +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 ### - @@ -1217,7 +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 ### @@ -1242,7 +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'"