address: 192.25.206.1
parents: gw-ubcece
hostgroups: layer3-infrastructure
+ gw-gatech:
+ address: 128.61.240.1
+ parents: gw-ubcece
+ hostgroups: layer3-infrastructure
gw-grnet:
address: 194.177.211.193
parents: gw-ubcece
address: 129.143.57.177
parents: gw-ubcece
hostgroups: layer3-infrastructure
+ gw-leaseweb:
+ address: 185.17.185.190
+ parents: gw-ubcece
+ hostgroups: layer3-infrastructure
gw-man-da:
address: 82.195.75.126
parents: gw-ubcece
chopin:
address: 195.20.242.124
parents: schumann
- hostgroups: computers, service, apache2-hosts, hassrvfs, hasbootfs, rsyncd-hosts, uploadqueue, kvmdomains, heavy-exim, xinetd-hosts, apache-https, postgres91-hosts, wheezy
+ hostgroups: computers, service, apache2-hosts, hassrvfs, hasbootfs, rsyncd-hosts, uploadqueue, kvmdomains, xinetd-hosts, apache-https, postgres91-hosts, wheezy
geo3:
address: 195.20.242.125
parents: schumann
abel:
address: 217.140.96.56
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, porterbox, wheezy, deadslow
- alain:
- address: 217.140.96.58
- parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, deadslow
- alwyn:
- address: 217.140.96.59
- parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, deadslow
+ hostgroups: computers, hasbootfs, hassrvfs, porterbox, wheezy, broken_mq
antheil:
address: 217.140.96.60
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, deadslow
- arne:
- address: 217.140.96.61
- parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, deadslow
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, broken_mq
arnold:
address: 217.140.96.57
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, deadslow
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, broken_mq
harris:
address: 217.140.96.66
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, porterbox, deadslow
+ hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, porterbox, broken_mq
+ hartmann:
+ address: 217.140.96.67
+ parents: gw-arm
+ hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, broken_mq
hasse:
address: 217.140.96.68
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, deadslow
+ hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, broken_mq
henze:
address: 217.140.96.70
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, deadslow
+ hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, broken_mq
hoiby:
address: 217.140.96.71
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd, deadslow
+ hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd, broken_mq
# }}}
# {{{ gw-brown
franck:
address: 138.16.160.12
parents: gw-brown
- hostgroups: computers, service, apache2-hosts, dl380, rsyncd-hosts, postgres91-hosts, spamd, heavy-exim, acpid-hosts, uploadqueue, xinetd-hosts, apache-https, hassrvfs, wheezy
+ hostgroups: computers, service, apache2-hosts, dl380, rsyncd-hosts, postgres91-hosts, spamd, acpid-hosts, uploadqueue, xinetd-hosts, apache-https, hassrvfs, wheezy
# }}}
# {{{ gw-bytemark
bm-bl1:
address: 5.153.231.248
parents: gw-bytemark
hostgroups: computers, bm-bl, acpid-hosts, service, wheezy
+ bm-bl9:
+ address: 5.153.231.249
+ parents: gw-bytemark
+ hostgroups: computers, bm-bl, acpid-hosts, service, wheezy, openstack-compute, broken_mq
+ bm-bl10:
+ address: 5.153.231.250
+ parents: gw-bytemark
+ hostgroups: computers, bm-bl, acpid-hosts, service, wheezy, openstack-compute, broken_mq
+ bm-bl11:
+ address: 5.153.231.251
+ parents: gw-bytemark
+ hostgroups: computers, bm-bl, acpid-hosts, service, wheezy, openstack-compute, broken_mq
+ bm-bl12:
+ address: 5.153.231.252
+ parents: gw-bytemark
+ hostgroups: computers, bm-bl, acpid-hosts, service, wheezy, openstack-compute, broken_mq
milanollo:
address: 5.153.231.2
address: 5.153.231.13
parents: ganeti-bytemark
hostgroups: computers, hassrvfs, kvmdomains, wheezy, apache2-hosts
- petrova:
- address: 5.153.231.25
- parents: ganeti-bytemark
- hostgroups: computers, kvmdomains, wheezy, apache2-hosts
couper:
address: 5.153.231.14
parents: ganeti-bytemark
hostgroups: computers, hassrvfs, kvmdomains, wheezy, apache2-hosts, nfs-client, autofs
rainier:
- address: 5.153.231.15
+ address: 5.153.231.16
parents: ganeti-bytemark
hostgroups: computers, kvmdomains, wheezy, no-bacula
rapoport:
- address: 5.153.231.16
+ address: 5.153.231.15
parents: ganeti-bytemark
hostgroups: computers, kvmdomains, wheezy, no-bacula
delfin:
address: 5.153.231.17
parents: ganeti-bytemark
- hostgroups: computers, hassrvfs, kvmdomains, wheezy, apache2-hosts
+ hostgroups: computers, hassrvfs, kvmdomains, wheezy, apache2-hosts, nfs-client, autofs
wuiet:
address: 5.153.231.18
parents: ganeti-bytemark
address: 5.153.231.20
parents: ganeti-bytemark
hostgroups: computers, general, kvmdomains, wheezy, nfs-client, autofs
+ moszumanska:
+ address: 5.153.231.21
+ parents: ganeti-bytemark
+ contact_groups: alioth-admins
+ hostgroups: computers, general, wheezy, postgres91-hosts, apache2-hosts, acpid-hosts, apache-https, brokensamhain, no-bacula, bind9-hosts, xinetd-hosts, alioth, heavy-exim, spamd
+ no-servicegroups: true
dillon:
address: 5.153.231.22
parents: ganeti-bytemark
address: 5.153.231.23
parents: ganeti-bytemark
hostgroups: computers, general, kvmdomains, wheezy, nfs-client, autofs, apache2-hosts, apache-https, service
- diamond:
- address: 5.153.231.24
+ petrova:
+ address: 5.153.231.25
+ parents: ganeti-bytemark
+ hostgroups: computers, kvmdomains, wheezy, apache2-hosts
+ oyens:
+ address: 5.153.231.26
+ parents: ganeti-bytemark
+ hostgroups: computers, kvmdomains, wheezy, apache2-hosts, openstack-controller, apache-https, broken_mq
+ barriere:
+ address: 5.153.231.27
+ parents: ganeti-bytemark
+ hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, porterbox
+ quantz:
+ address: 5.153.231.28
parents: ganeti-bytemark
- hostgroups: computers, service, kvmdomains, wheezy, bind9-hosts, no-bacula
+ hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, nfs-client, xinetd-hosts, heavy-exim, apache2-hosts, autofs, apache-https
+ portman:
+ address: 5.153.231.29
+ parents: ganeti-bytemark
+ hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, apache2-hosts
+ paradis:
+ address: 5.153.231.30
+ parents: ganeti-bytemark
+ hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, apache2-hosts
# }}}
# {{{ gw-c3sl
santoro:
senfl:
address: 128.31.0.51
parents: gw-csail
- hostgroups: computers, service, dl360, acpid-hosts, hassrvfs, apache2-hosts, rsyncd-hosts, bind9-hosts, xinetd-hosts, squeeze, apache-https
+ hostgroups: computers, service, dl360, acpid-hosts, hassrvfs, apache2-hosts, rsyncd-hosts, xinetd-hosts, squeeze, apache-https
steffani:
address: 128.31.0.36
parents: gw-csail
hostgroups: computers, service, apache2-hosts, rsyncd-hosts, sw-raid, acpid-hosts, hasbootfs, hasorgfs, xinetd-hosts, wheezy, security_mirror, no-bacula
# }}}
# {{{ gw-dgi
- argento:
- address: 93.94.130.160
+ storace:
+ address: 93.94.130.161
parents: gw-dgi
- hostgroups: computers, sw-raid, hassrvfs, wheezy
+ hostgroups: computers, acpid-hosts, wheezy, dl380, nfs-client, hassrvfs
# }}}
# {{{ gw-ftcollins
#alkman:
# address: 192.25.206.63
# parents: gw-ftcollins
# hostgroups: computers, buildd, acpid-hosts, wheezy
- #merulo:
- # address: 192.25.206.58
- # parents: gw-ftcollins
- # hostgroups: computers, porterbox, hasusrfs, wheezy
#mundy:
# address: 192.25.206.62
# parents: gw-ftcollins
parents: gw-ftcollins
hostgroups: computers, service, dl380, apache2-hosts, wheezy, no-bacula
# }}}
+ # {{{ gw-gatech
+ sechter:
+ address: 128.61.240.73
+ parents: gw-gatech
+ hostgroups: computers, service, apache2-hosts, rsyncd-hosts, sw-raid, acpid-hosts, hasbootfs, hassrvfs, xinetd-hosts, wheezy, security_mirror
+ # }}}
# {{{ gw-grnet
barber:
address: 194.177.211.203
address: 194.177.211.204
parents: gw-grnet
hostgroups: computers, acpid-hosts, buildd, hassrvfs, mptraid, wheezy
- cilea:
- address: 194.177.211.205
- parents: gw-grnet
- hostgroups: computers, acpid-hosts, mptraid, hassrvfs, service, squeeze
- orff:
- address: 194.177.211.209
- parents: gw-grnet
- hostgroups: computers, acpid-hosts, bind9-hosts, mptraid, service, hassrvfs, apache2-hosts, squeeze
rautavaara:
address: 194.177.211.199
parents: gw-grnet
parents: gw-isc
hostgroups: computers, service, apache2-hosts, rsyncd-hosts, acpid-hosts, dl360, hasorgfs, xinetd-hosts, wheezy, security_mirror, no-bacula
# }}}
+ # {{{ gw-leaseweb
+ lw01:
+ address: 185.17.185.177
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl180, nfs-server
+ lw02:
+ address: 185.17.185.178
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl180, nfs-server
+ lw03:
+ address: 185.17.185.179
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl180, nfs-server
+ lw04:
+ address: 185.17.185.180
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl180, nfs-server, postgres91-hosts
+ lw05:
+ address: 185.17.185.181
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl120, sw-raid, nfs-client, autofs, hassrvfs, apache2-hosts
+ lw06:
+ address: 185.17.185.182
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl120, sw-raid, nfs-client, autofs, hassrvfs, apache2-hosts
+ # }}}
# {{{ gw-karlsruhe
zemlinsky:
address: 129.143.160.6
parents: gw-karlsruhe
- hostgroups: computers, buildd, hassrvfs, wheezy
+ hostgroups: computers, buildd, wheezy
contacts: pkern
# }}}
# {{{ gw-man-da
address: 82.195.75.112
parents: ganeti3
hostgroups: computers, service, hassrvfs, apache2-hosts, kvmdomains, wheezy, acpid-hosts, apache-https, rsyncd-hosts, xinetd-hosts
- vieuxtemps:
- address: 82.195.75.113
- parents: ganeti3
- hostgroups: computers, service, kvmdomains, wheezy, varnish-hosts
mailly:
address: 82.195.75.114
parents: ganeti3
zani:
address: 148.100.88.22
parents: gw-marist
- hostgroups: computers, buildd, hassrvfs, wheezy, incomingmailrelayed
+ hostgroups: computers, buildd, hassrvfs, wheezy, incomingmailrelayed, ping-suckers
# }}}
# {{{ gw-osuosl
busoni:
# address: 140.211.166.58
# parents: gw-osuosl
# hostgroups: computers, buildd, hasbootfs
+ merulo:
+ address: 140.211.166.46
+ parents: gw-osuosl
+ hostgroups: computers, porterbox, hasusrfs, wheezy
parry:
address: 140.211.15.153
parents: gw-osuosl
ravel:
address: 206.12.19.5
parents: sw-ubcece-kais
- hostgroups: computers, general, dl385, apache2-hosts, acpid-hosts, hasbootfs, nfs-client, rsyncd-hosts, bind9-hosts, uploadqueue, hasorgfs, heavy-exim, xinetd-hosts, wheezy, autofs
+ hostgroups: computers, general, dl385, apache2-hosts, acpid-hosts, hasbootfs, nfs-client, rsyncd-hosts, uploadqueue, hasorgfs, xinetd-hosts, wheezy, autofs
dijkstra:
address: 206.12.19.218
parents: sw-ubcece-kais
address: 206.12.19.13
parents: sw-ubcece-kais
hostgroups: computers, hashomefs, sw-raid, rsyncd-hosts, apache2-hosts, xinetd-hosts, service, nfs-server, squeeze, hassrvfs
- respighi:
- address: 206.12.19.11
- parents: sw-ubcece-kais
- hostgroups: computers, hasbootfs, aacraid, hassrvfs, service, apache2-hosts, squeeze
# MSA 2000 (2012i)
giustini:
address: 192.168.2.6
address: 206.12.19.121
parents: sw-ubcece-kais
hostgroups: computers, service, hasbootfs, kvmdomains, wheezy, hassrvfs
- quantz:
- address: 206.12.19.122
- parents: traetta
- hostgroups: computers, service, hasbootfs, kvmdomains, wheezy, hassrvfs, nfs-client, xinetd-hosts, heavy-exim, apache2-hosts, autofs
nono:
address: 206.12.19.123
parents: traetta
- hostgroups: computers, service, kvmdomains, wheezy, heavy-exim, xinetd-hosts, apache2-hosts, apache-https
+ hostgroups: computers, service, kvmdomains, wheezy, heavy-exim, xinetd-hosts, apache2-hosts, apache-https, broken_https_default_vhost
reger:
address: 206.12.19.124
parents: ganeti2
gombert:
address: 206.12.19.132
parents: ganeti2
- hostgroups: computers, service, kvmdomains, wheezy, heavy-exim, apache2-hosts
+ hostgroups: computers, service, kvmdomains, wheezy, apache2-hosts
jenkins:
address: 206.12.19.133
parents: salieri
address: 206.12.19.134
parents: ganeti2
hostgroups: computers, service, kvmdomains, wheezy, xinetd-hosts, nfs-client, autofs
- barriere:
- address: 206.12.19.135
- parents: ganeti2
- hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, porterbox
diabelli:
address: 206.12.19.136
parents: traetta
- hostgroups: computers, service, hasbootfs, kvmdomains, wheezy, apache2-hosts, apache-https
+ hostgroups: computers, service, hasbootfs, kvmdomains, wheezy, apache2-hosts, apache-https, broken_https_default_vhost
bizet:
address: 206.12.19.137
parents: ganeti2
hostgroups: secondary-IPs
# }}}
# {{{ gw-ynic
- hildegard:
- address: 144.32.168.74
- parents: gw-ynic
- hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, deadslow, buildd
howells:
address: 144.32.168.75
parents: gw-ynic
- hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, deadslow, buildd
+ hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd
hummel:
address: 144.32.168.76
parents: gw-ynic
- hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, deadslow, buildd
+ hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd
porpora:
address: 144.32.168.78
parents: gw-ynic
# }}}
# }}}
-
-#############################
-# host groups
-#
-# hostgroups ircd and all are automatically defined
-#
-#############################
+# {{{ ############################# host groups #############################
hostgroups:
computers:
alias: computers
notacomputer:
alias: Systems that are not really systems. Yeah :)
private: 1
- deadslow:
- alias: Systems too slow to run any real checks
freebsd:
alias: freebsd
private: 1
dl585:
alias: HP DL385 hosts
private: 1
+ dl180:
+ alias: HP DL180
+ private: 1
+ dl120:
+ alias: HP DL120
+ private: 1
sw-raid:
alias: Hosts with Linux software raid
private: 1
aacraid:
alias: Hosts with Adaptec AACraid
private: 1
-# megactl:
-# alias: Hosts with LSI Logic MegaRAID, but not usable with megaraid check
-# private: 1
megaraid:
alias: Hosts with LSI Logic MegaRAID
private: 1
jetty-hosts:
alias: hosts running jetty
private: 1
- varnish-hosts:
- alias: hosts running varnish
- private: 1
bind9-hosts:
alias: hosts running bind9
private: 1
# once every host runs unbound, do away with this group and check "computers" instead
- unbound-hosts:
- alias: hosts running unbound
- private: 1
amavis-hosts:
alias: hosts running amavis
private: 1
- #tftpd-hosts:
- # alias: hosts running a tftpd (tftpd-hpa as a daemon)
- # private: 1
rsyncd-hosts:
alias: hosts providing rsync services via xinetd
private: 1
xinetd-hosts:
alias: hosts providing services via xinetd
private: 1
- #postgres81-hosts:
- # alias: hosts running postgres81
- # private: 1
postgres91-hosts:
alias: hosts running postgres91
private: 1
apache-https:
alias: hosts with https services
private: 1
+ broken_https_default_vhost:
+ alias: https default vhost does not say 200 OK
+ private: 1
no-bacula:
alias: hosts which are not being backed up with bacula
hasusrfs:
alias: hosts with a /usr filesystem
private: 1
-# hasvicepa:
-# alias: hosts with a /vicepa
-# private: 1
incomingmailrelayed:
alias: incoming mail needs to go through a mail relay
# i.e. no port 25
private: 1
- ntpsuckers:
- alias: "hosts who's ntp offset is often unknown"
- private: 1
-
brokensamhain:
alias: machines that can not run samhain
private: 1
high-RTT:
- alias: machines with hight round trip times
+ alias: machines with high round trip times
+ private: 1
+ ping-suckers:
+ alias: machines that just suck at icmp
+ private: 1
+ alioth:
+ alias: machines that just are just awkward
+ private: 1
+ openstack-compute:
+ alias: nodes that run OpenStack compute
+ private: 1
+ openstack-controller:
+ alias: nodes that run OpenStack controller
private: 1
security_mirror:
alias: hosts that are security mirrors
private: 1
-#############################
-# servicegroups
-#############################
+
+ broken_mq:
+ alias: hosts whose MQ is broken
+ private: 1
+# }}}
+# {{{ ############################# servicegroups #############################
servicegroups:
diskspace:
alias: diskusage checks
security:
alias: security
servicegroup_members: apt, kernel, samhain
-
-#############################
-# services
-#############################
+ MQ:
+ alias: rabbitMQ stuff
+# }}}
+# {{{ ############################# services #############################
services:
+ # {{{ ### basic networking
-
name: PING
check: "check_ping!350.0,20%!600.0,40%"
hostgroups: pingable
- excludehostgroups: layer3-infrastructure, high-RTT
+ excludehostgroups: layer3-infrastructure, high-RTT, ping-suckers
normal_check_interval: 5
max_check_attempts: 4
retry_check_interval: 1
normal_check_interval: 5
max_check_attempts: 4
retry_check_interval: 1
+ -
+ name: PING
+ check: "check_ping!600.0,90%!900.0,95%"
+ hostgroups: ping-suckers
+ normal_check_interval: 5
+ max_check_attempts: 4
+ retry_check_interval: 1
-
name: PING
check: "check_ping!2000.0,60%!3000.0,80%"
normal_check_interval: 5
max_check_attempts: 4
retry_check_interval: 1
-
- ############ Services ############
- ###
-
- ############ Disk Usage ############
- ####
-
+ # }}}
+ # {{{ ### disk usage
-
name: disk usage - all
servicegroups: diskspace
name: disk usage on /var/lib/postgresql
servicegroups: diskspace
nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /var/lib/postgresql"
- hosts: sibelius, busoni, buxtehude
+ hosts: sibelius, busoni, buxtehude, lw04
-
name: disk usage on /var/log
servicegroups: diskspace
servicegroups: diskspace
nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/ftp-master.debian.org"
hosts: franck
- -
- name: disk usage on /srv/pgbackup
- servicegroups: diskspace
- nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /srv/pgbackup"
- hosts: franck
-
name: disk usage on /srv/pgbackup
servicegroups: diskspace
servicegroups: diskspace
nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/bacula"
hosts: beethoven
-
- ############ All Computers ############
- ####
-
- name: apt - security updates
- servicegroups: apt
- nrpe: "/usr/lib/nagios/plugins/dsa-check-statusfile /var/cache/dsa/nagios/apt"
+ name: disk usage on /storage/snapshot-farm-1
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 98 92 /storage/snapshot-farm-1"
+ hosts: lw01
+ -
+ name: disk usage on /storage/snapshot-farm-2
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 98 92 /storage/snapshot-farm-2"
+ hosts: lw02
+ -
+ name: disk usage on /storage/snapshot-farm-3
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 98 92 /storage/snapshot-farm-3"
+ hosts: lw03
+ -
+ name: disk usage on /storage/snapshot-farm-4
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 98 92 /storage/snapshot-farm-4"
+ hosts: lw04
+ -
+ name: disk usage on /srv/morgue.debian.org/
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 95 90 /srv/morgue.debian.org"
+ hosts: lw03
+ # }}}
+ # {{{ ### system
+ # {{{ setup
+ -
+ name: setup - dsa config
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-config"
+ hostgroups: computers
+ normal_check_interval: 60
+ excludehostgroups: alioth
+ -
+ name: setup - local hostname etc-hosts
+ nrpe: 'if getent ahosts `hostname` | grep -q 127.0; then echo "Warning: local hostname resolves to 127/8 address"; exit 1; else echo "OK: Hostname resolves to non-127/8 address."; exit 0; fi'
+ hostgroups: computers
+ normal_check_interval: 60
+ # }}}
+ # {{{ os health
+ ####
+ -
+ name: users
+ nrpe: "/usr/lib/nagios/plugins/check_users 30 35"
+ hostgroups: computers
+ -
+ name: load
+ nrpe: "/usr/lib/nagios/plugins/check_load -w 30,28,26 -c 50,45,50"
+ hostgroups: computers
+ excludehostgroups: highload
+ -
+ name: load
+ nrpe: "/usr/lib/nagios/plugins/check_load -w 140,120,100 -c 240,220,200"
+ hostgroups: highload
+ -
+ name: uptime check
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-uptime"
+ hostgroups: computers
+ -
+ name: processes - total
+ nrpe: "/usr/lib/nagios/plugins/check_procs 620 700"
+ hostgroups: computers
+ -
+ name: swap usage - percent
+ nrpe: "/usr/lib/nagios/plugins/check_swap -w 20% -c 10%"
+ hostgroups: computers
+ -
+ name: swap usage - mb
+ nrpe: "/usr/lib/nagios/plugins/check_swap -w 20000 -c 5000"
+ hostgroups: computers
+ -
+ name: process - getty
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C getty -a /sbin/getty"
+ hostgroups: computers
+ excludehosts: zelenka, zandonai
+ -
+ name: processes - zombies
+ nrpe: "/usr/lib/nagios/plugins/check_procs 5 10 -s Z"
+ hostgroups: computers
+ -
+ name: system - available entropy
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-entropy"
+ event_handler: dsa_event_handler_restart_ekey
hostgroups: computers
+ excludehostgroups: freebsd
+ -
+ name: system - filesystem check
+ nrpe: "/usr/bin/sudo /usr/lib/nagios/plugins/dsa-check-filesystems"
normal_check_interval: 60
retry_check_interval: 15
- ####
+ hostgroups: computers
+ # }}}
+ # {{{ backup
-
name: backup
servicegroups: backup
servicegroups: backup
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u bacula -C bacula-fd -a '/usr/sbin/bacula-fd -c /etc/bacula/bacula-fd.conf'"
hostgroups: computers
- excludehostgroups: freebsd
+ excludehostgroups: freebsd, alioth
-
name: process - bacula-fd
servicegroups: backup
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u root -C bacula-fd -a '/usr/sbin/bacula-fd -c /etc/bacula/bacula-fd.conf'"
hostgroups: freebsd
- ####
+ ####
+ -
+ name: process - acc.umu.se backup
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -a 'dsmc'"
+ hosts: sibelius
+ # }}}
+ # {{{ security
-
name: running kernel
servicegroups: kernel
hostgroups: computers
normal_check_interval: 60
retry_check_interval: 5
-
- ####
-# - name: afs - space
-# hostgroups: bosserver
-# nrpe: "/usr/lib/nagios/plugins/dsa-check-afs-space -H localhost"
-# normal_check_interval: 60
-# max_check_attempts: 2
-# retry_check_interval: 5
-#
-# - name: afs - bos
-# hostgroups: bosserver
-# nrpe: "/usr/lib/nagios/plugins/dsa-check-afs-bos -H localhost"
-# normal_check_interval: 60
-# max_check_attempts: 2
-# retry_check_interval: 5
-#
-# - name: afs - waiting connections
-# hostgroups: bosserver
-# nrpe: "/usr/lib/nagios/plugins/dsa-check-afs-rxdebug -H localhost"
-# normal_check_interval: 60
-# max_check_attempts: 2
-# retry_check_interval: 5
-#
- #- name: afs - udebug
- # hostgroups: bosserver
- # nrpe: "/usr/lib/nagios/plugins/dsa-check-afs-udebug -H localhost"
- # normal_check_interval: 60
- # max_check_attempts: 2
- # retry_check_interval: 5
-
- ####
- #-
- # name: puppet
- # nrpe: "/usr/lib/nagios/plugins/dsa-check-file_age -i 540 -f /var/lib/puppet/state/state.yaml"
- # hostgroups: computers
-
- ####
+ -
+ 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: installed firewall
nrpe: "/usr/lib/nagios/plugins/dsa-check-file -w -f /etc/ferm/ferm.conf"
nrpe: "/usr/lib/nagios/plugins/dsa-check-file -w -f /etc/ferm/conf.d/defs.conf"
hostgroups: computers
excludehostgroups: freebsd
- ####
- - name: ganeti - job watcher paused
- nrpe: "/usr/lib/nagios/plugins/negate /usr/lib/nagios/plugins/dsa-check-file -f /var/lib/ganeti/watcher.pause"
+ -
+ name: process - ulogd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C ulogd -a '/usr/sbin/ulogd -d'"
hostgroups: computers
- ####
+ excludehostgroups: freebsd, sparc
+ -
+ name: unexpected process - ulogd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C ulogd"
+ hostgroups: freebsd, sparc
+ ####
-
name: process - samhain
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C samhain -a '/usr/sbin/samhain'"
normal_check_interval: 60
retry_check_interval: 5
excludehostgroups: brokensamhain
- ####
- -
- name: process - acc.umu.se backup
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -a 'dsmc'"
- hosts: sibelius
- ####
- -
- name: users
- nrpe: "/usr/lib/nagios/plugins/check_users 30 35"
- hostgroups: computers
- ####
- -
- name: load
- nrpe: "/usr/lib/nagios/plugins/check_load -w 30,28,26 -c 50,45,50"
- hostgroups: computers
- excludehostgroups: highload
- -
- name: load
- nrpe: "/usr/lib/nagios/plugins/check_load -w 140,120,100 -c 240,220,200"
- hostgroups: highload
- ####
- -
- name: uptime check
- nrpe: "/usr/lib/nagios/plugins/dsa-check-uptime"
- hostgroups: computers
- ####
-
name: processes - samhain zombies
nrpe: "/usr/lib/nagios/plugins/check_procs 3 6 -s Z -u root -a samhain"
event_handler: dsa_event_handler_restart_samhain
hostgroups: computers
excludehostgroups: brokensamhain
+ # }}}
+ # {{{ logging
-
- name: processes - zombies
- nrpe: "/usr/lib/nagios/plugins/check_procs 5 10 -s Z"
- hostgroups: computers
- ####
- -
- name: processes - total
- nrpe: "/usr/lib/nagios/plugins/check_procs 620 700"
+ 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: computers
- ####
+ excludehostgroups: freebsd
+
-
- name: swap usage - percent
- nrpe: "/usr/lib/nagios/plugins/check_swap -w 20% -c 10%"
- hostgroups: computers
- ####
+ name: process - syslog-ng
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 2: -u root -C syslog-ng -a '/sbin/syslog-ng -p /var/run/syslog-ng.pid'"
+ hostgroups: freebsd
+
-
- name: swap usage - mb
- nrpe: "/usr/lib/nagios/plugins/check_swap -w 20000 -c 5000"
+ name: remote logging on lotti
+ remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
+ runfrom: lotti
hostgroups: computers
- ####
+ excludehostgroups: alioth
-
- name: process - getty
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C getty -a /sbin/getty"
+ name: remote logging on lully
+ remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
+ runfrom: lully
hostgroups: computers
- excludehosts: zelenka, zandonai
- ####
+ excludehostgroups: alioth
+ # }}}
+ # {{{ base service
-
name: process - sshd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:20 -c 1: -u root -C sshd -a '/usr/sbin/sshd'"
depends: process - sshd
normal_check_interval: 60
notification_interval: 1440
- ####
+ ####
-
name: network service - nrpe
check: check_tcp!5666
hostgroups: computers
max_check_attempts: -1
depends: network service - nrpe
- ###
+ ###
-
name: process - munin-node
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u root -C munin-node -a '/usr/sbin/munin-node'"
check: check_tcp!4949
hostgroups: computers
depends: process - munin-node
- ###
+ ###
-
name: process - ntpd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -p 1 -C ntpd -a '/usr/sbin/ntpd -p /var/run/ntpd.pid'"
hostgroups: computers
depends: process - ntpd
excludehosts: ancina
- excludehostgroups: ntpsuckers, deadslow
servicegroups: time
- #
+ ###
-
name: network service - time
check: dsa_check_time
hosts: ancina
depends: process - xinetd
servicegroups: time
-
- ###
+ ###
-
name: process - atd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u daemon -C atd -a /usr/sbin/atd"
hostgroups: computers
- ###
+ ###
-
name: process - cron
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u root -C cron -a /usr/sbin/cron"
hostgroups: computers
-
- ###
+ ###
-
- 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'"
+ name: process - ud-replicated
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C ud-replicated -a '/usr/bin/python /usr/bin/ud-replicated'"
hostgroups: computers
- excludehostgroups: freebsd
-
+ excludehostgroups: squeeze, freebsd, alioth
-
- name: process - syslog-ng
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 2: -u root -C syslog-ng -a '/sbin/syslog-ng -p /var/run/syslog-ng.pid'"
+ name: process - ud-replicated
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C python2.7 -a '/usr/bin/python /usr/bin/ud-replicated'"
hostgroups: freebsd
-
+ ###
-
- name: remote logging on lotti
- remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
- runfrom: lotti
- hostgroups: computers
+ name: process - monit
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/sbin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
+ hostgroups: squeeze
-
- name: remote logging on lully
- remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
- runfrom: lully
+ name: process - monit
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/bin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
hostgroups: computers
+ excludehostgroups: squeeze, alioth
+ ###
-
name: MQ connection on rainier
+ servicegroups: MQ
remotecheck: "/usr/lib/nagios/plugins/dsa-check-mq-connection $HOSTNAME$ ud dsa"
runfrom: rainier
hostgroups: computers
normal_check_interval: 60
retry_check_interval: 15
+ excludehostgroups: alioth, broken_mq
-
name: MQ connection on rapoport
+ servicegroups: MQ
remotecheck: "/usr/lib/nagios/plugins/dsa-check-mq-connection $HOSTNAME$ ud dsa"
runfrom: rapoport
hostgroups: computers
normal_check_interval: 60
retry_check_interval: 15
- ### MAIL STUFF
- ###
+ excludehostgroups: alioth, broken_mq
+ ###
-
- name: process - exim
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
+ name: local resolver
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-resolver www.debian.org www.google.com"
hostgroups: computers
- excludehostgroups: postfix-hosts, mail-relay
- excludehosts: master, busoni, quantz, buxtehude
- -
- name: process - exim
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
- hostgroups: mail-relay
+ normal_check_interval: 60
-
- name: process - exim - total
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1: -C exim4"
+ name: process - unbound
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u unbound -C unbound -a '/usr/sbin/unbound'"
hostgroups: computers
- excludehostgroups: postfix-hosts
- excludehosts: master, busoni, quantz, buxtehude
- -
- name: process - exim
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:300 -c 1:500 -C exim4 -a '/usr/sbin/exim4'"
- hosts: master, busoni, quantz, buxtehude
+ excludehostgroups: alioth
+ ###
-
- name: mail queue
- nrpe: "/usr/lib/nagios/plugins/check_mailq -M exim -w 1000 -c 2000"
- hostgroups: heavy-exim
- ###
- -
- name: process - clamav - clamd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u clamav -C clamd -a '/usr/sbin/clamd'"
- hostgroups: heavy-exim, heavy-postfix
- -
- name: service - clamav
- nrpe: "/usr/lib/nagios/plugins/check_clamd -H /var/run/clamav/clamd.ctl"
- hostgroups: heavy-exim, heavy-postfix
- depends: process - clamav - clamd
- -
- 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: unwanted process - clamav
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C clamd"
+ name: process - uptimed
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u daemon -C uptimed -a '/usr/sbin/uptimed'"
hostgroups: computers
- excludehostgroups: heavy-exim, heavy-postfix, deadslow
+ ###
-
- name: unwanted process - freshclam
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C freshclam"
+ name: process - udevd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -p 1 -C udevd -a 'udevd'"
hostgroups: computers
- excludehostgroups: heavy-exim, heavy-postfix, deadslow
- ###
- -
- name: process - spamd - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd --create-prefs --max-children 5 --helper-home-dir -d --pidfile=/var/run/spamd.pid'"
- hostgroups: spamd
- excludehosts: picconi
- excludehostgroups: deadslow
- -
- name: process - spamd - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd --create-prefs --max-children 20 --min-spare=5 --helper-home-dir -d --pidfile=/var/run/spamd.pid'"
- hosts: picconi
- -
- name: process - spamd - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd --create-prefs --max-children 10 --helper-home-dir -d --pidfile=/var/run/spamd.pid'"
- hosts: bendel
- -
- name: process - spamd - child
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:11 -c 1: -C spamd -a 'spamd child'"
- hosts: bendel
- hostgroups: spamd
- depends: process - spamd - master
- #
- -
- name: process - spamd - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u debbugs -C spamd -a '/usr/sbin/spamd -d '"
- hosts: buxtehude
+ excludehostgroups: freebsd
-
- name: process - spamd - child
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1: -u debbugs -C spamd -a 'spamd child'"
- hosts: buxtehude
- #
+ name: unexpected process - udev
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C udevd"
+ hostgroups: freebsd
+ ###
-
- name: unwanted process - spamd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd"
- hostgroups: computers
- excludehostgroups: spamd, deadslow
- excludehosts: bendel, busoni, buxtehude
-
- ###
- #-
- # name: process - greylistd
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u greylist -C greylistd -a '/usr/bin/python /usr/sbin/greylistd'"
- # hostgroups: heavy-exim
- #
+ name: process - acpid
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C acpid -a '/usr/sbin/acpid'"
+ hostgroups: acpid-hosts
-
- name: unwanted process - greylistd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C greylistd"
+ name: unexpected process - acpid
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C acpid"
hostgroups: computers
- excludehostgroups: deadslow
-
- ###
- -
- name: process - postgrey
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --unix=/var/run/postgrey/socket --retry-window=4 --auto-whitelist-clients=10 --exim'"
- hostgroups: heavy-exim
+ excludehostgroups: acpid-hosts, kvmdomains
+ ###
-
- name: process - postgrey
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --inet=127.0.0.1:60000'"
- hostgroups: heavy-postfix
- #
+ name: process - xinetd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xinetd -a '/usr/sbin/xinetd -pidfile /var/run/xinetd.pid -stayalive'"
+ hostgroups: xinetd-hosts
-
- name: unwanted process - postgrey
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C postgrey"
+ name: unwanted process - xinetd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C xinetd"
hostgroups: computers
- excludehostgroups: heavy-postfix, heavy-exim, deadslow
- ###
- -
- name: process - amavis - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u amavis -a 'amavisd-new (master)'"
- hostgroups: amavis-hosts
+ excludehostgroups: xinetd-hosts
+ ###
-
- name: process - amavis - all
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1:10 -u amavis -a 'amavisd-new '"
- hostgroups: amavis-hosts
- depends: process - amavis - master
- #
+ name: process - stunnel4 - puppet-ekeyd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1: -u stunnel4 -C stunnel4 -a '/usr/bin/stunnel4 /etc/stunnel/puppet-ekeyd.conf'"
+ hostgroups: squeeze, wheezy
+ excludehostgroups: freebsd, alioth
+ # }}}
+ # {{{ anti-services
-
- name: unwanted process - amavis
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C amavisd-new"
+ name: ganeti - job watcher paused
+ nrpe: "/usr/lib/nagios/plugins/negate /usr/lib/nagios/plugins/dsa-check-file -f /var/lib/ganeti/watcher.pause"
hostgroups: computers
- excludehostgroups: amavis-hosts, deadslow
- ###
- -
- name: process - weightd - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (master)'"
- hostgroups: heavy-postfix
- -
- name: process - weightd - cache
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (cache)'"
- hostgroups: heavy-postfix
- depends: process - weightd - master
- -
- name: process - weightd - child
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:50 -c 1: -u polw -a 'policyd-weight (child)'"
- hostgroups: heavy-postfix
- depends: process - weightd - master
- ###
-
- name: unwanted process - policyd-weight
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C policyd-weight"
+ name: unwanted process - irqbalance
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance"
hostgroups: computers
- excludehostgroups: heavy-postfix, deadslow
-
-
- ###
- -
- name: process - postfix - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C master -a '/usr/lib/postfix/master'"
- hostgroups: postfix-hosts
- -
- name: process - postfix - qmgr
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C qmgr -a 'qmgr -l -t fifo -u'"
- hostgroups: postfix-hosts
- depends: process - postfix - master
- #-
- # name: process - postfix - tlsmgr
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C tlsmgr -a 'tlsmgr -l -t fifo -u'"
- # hostgroups: postfix-hosts
- # depends: process - postfix - master
- -
- name: process - postfix - pickup
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C pickup -a 'pickup -l -t fifo -u -c'"
- hostgroups: postfix-hosts
- depends: process - postfix - master
- -
- name: process - postfix - anvil
- 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
-
- -
- name: process - postfix - trivial-rewrite
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C trivial-rewrite -a 'trivial-rewrite -n rewrite -t unix -u -c'"
- hostgroups: postfix-hosts
- depends: process - postfix - master
- -
- name: process - postfix - proxymap
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C proxymap -a 'proxymap -t unix -u'"
- hostgroups: postfix-hosts
- depends: process - postfix - master
- -
- name: process - postfix - cleanup
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u postfix -C cleanup -a 'cleanup -z -t unix -u -c'"
- hostgroups: postfix-hosts
- depends: process - postfix - master
- -
- name: process - postfix - local
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u postfix -C local -a 'local -t unix'"
- hostgroups: postfix-hosts
- depends: process - postfix - master
-
name: unwanted process - openvpn
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C openvpn"
hostgroups: computers
normal_check_interval: 120
-
-
- ###
- -
- name: network service - smtp
- check: dsa_check_smtp
- hostgroups: computers
- excludehostgroups: postfix-hosts, incomingmailrelayed, incomingmailrelayed2025
- depends: process - exim
-
- -
- name: network service - smtp
- check: dsa_check_smtp
- hostgroups: postfix-hosts
- depends: process - postfix - master
- -
- name: network service - submission
- check: dsa_check_smtp_port!587
- hostgroups: incomingmailrelayed
- depends: process - exim
- -
- name: network service - smtp 2025
- check: dsa_check_smtp_port!2025
- hostgroups: incomingmailrelayed2025
- depends: process - exim
- -
- name: network service local - smtps cert
- 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: 120
-
- -
- name: setup - dsa config
- nrpe: "/usr/lib/nagios/plugins/dsa-check-config"
- hostgroups: computers
- normal_check_interval: 60
- -
- name: setup - local hostname etc-hosts
- nrpe: 'if getent ahosts `hostname` | grep -q 127.0; then echo "Warning: local hostname resolves to 127/8 address"; exit 1; else echo "OK: Hostname resolves to non-127/8 address."; exit 0; fi'
- hostgroups: computers
- normal_check_interval: 60
- -
- name: system - available entropy
- nrpe: "/usr/lib/nagios/plugins/dsa-check-entropy"
- event_handler: dsa_event_handler_restart_ekey
- hostgroups: computers
- excludehostgroups: freebsd
- -
- name: system - filesystem check
- nrpe: "/usr/bin/sudo /usr/lib/nagios/plugins/dsa-check-filesystems"
- normal_check_interval: 60
- retry_check_interval: 15
- hostgroups: computers
- ###
- -
- name: local resolver
- nrpe: "/usr/lib/nagios/plugins/dsa-check-resolver www.debian.org www.google.com"
- hostgroups: computers
- normal_check_interval: 60
- -
- name: process - unbound
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u unbound -C unbound -a '/usr/sbin/unbound'"
- hostgroups: unbound-hosts, squeeze, wheezy
- ###
- -
- name: process - uptimed
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u daemon -C uptimed -a '/usr/sbin/uptimed'"
- hostgroups: computers
- ###
- -
- name: unwanted process - irqbalance
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance"
- hostgroups: computers
- excludehostgroups: deadslow
-
- ####
- ###
- #-
- # name: unwanted process - system-tools-backends
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C system-tools-ba"
- # hostgroups: computers
- #-
- # name: unwanted process - dbus-daemon
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C dbus-daemon"
- # hostgroups: computers
-
name: unwanted process - gkrellmd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C gkrellmd"
hostgroups: computers
- excludehostgroups: deadslow
-
name: unwanted process - rpc.statd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C rpc.statd"
hostgroups: computers
- excludehostgroups: nfs-client, nfs-server, deadslow
+ excludehostgroups: nfs-client, nfs-server
-
name: unwanted process - inetd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C inetd"
hostgroups: computers
- excludehosts: abel, alwyn
- excludehostgroups: deadslow
+ excludehosts: abel
-
name: unwanted process - snmpd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C snmpd"
hostgroups: computers
- excludehostgroups: deadslow
-
- ####
+ # }}}
+ # {{{ ssl certs
-
name: "host SSL cert"
nrpe: "if [ -e /etc/ssl/certs/thishost.pem ]; then /usr/lib/nagios/plugins/dsa-check-cert-expire /etc/ssl/certs/thishost.pem; else echo 'No thishost.pem on this host.'; fi"
hostgroups: computers
-
- ############ Processes/Services that only run on some computers ############
- ####
- ###
- -
- 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'"
- hostgroups: dl385
- ###
+ # }}}
+ # {{{ HW health/raid
-
name: process - mdadm monitor
servicegroups: raid
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-sw"
hostgroups: sw-raid
-
- ###
- -
- name: process - ud-replicated
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C ud-replicated -a '/usr/bin/python /usr/bin/ud-replicated'"
- hostgroups: computers
- excludehostgroups: squeeze,freebsd
- -
- name: process - ud-replicated
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C python2.7 -a '/usr/bin/python /usr/bin/ud-replicated'"
- hostgroups: freebsd
- -
- name: process - monit
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/sbin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
- hostgroups: squeeze
- -
- name: process - monit
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/bin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
- hostgroups: computers
- excludehostgroups: squeeze
+ ###
-
name: HW - hpacucli status
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpacucli"
normal_check_interval: 120
- hostgroups: dl385, dl380, dl360, bl460
+ hostgroups: dl385, dl380, dl360, bl460, dl180
excludehosts: schein, rietz
-
name: HW - hpacucli status
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpacucli --ignore-transfer-speed=1I:1:1 --ignore-transfer-speed=1I:1:2"
normal_check_interval: 120
hostgroups: dl585
- ###
+ ###
+ -
+ name: HW - edac status
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-edac"
+ normal_check_interval: 120
+ hosts: lw05, lw06
+ #hostgroups: computers
+ #excludehosts: villa, lobos, senfl, schein
-
name: HW - hpasmcli status
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm"
normal_check_interval: 120
hostgroups: dl385, dl380, dl360, bl460, dl585, bm-bl
- excludehosts: villa, lobos, senfl, schein
+ excludehosts: villa, lobos, senfl, schein, storace
-
name: HW - hpasmcli status
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm --ps-no-redundant"
normal_check_interval: 120
- hosts: villa, lobos
+ hosts: villa
+ -
+ name: HW - hpasmcli status
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm --ps-no-redundant --ignore-failed='PS2'"
+ normal_check_interval: 120
+ hosts: lobos
-
name: HW - hpasmcli status
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm --fan-no-redundant"
normal_check_interval: 120
hosts: senfl, schein
- ###
+ -
+ name: HW - hpasmcli status
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm --fan-ignore-not-present"
+ normal_check_interval: 120
+ hosts: storace
+ ###
-
name: RAID - 3ware
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-3ware"
hosts: powell
- ###
+ ###
-
name: RAID - aacraid
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-aacraid"
hostgroups: aacraid
- ###
+ ###
-
name: RAID - megaraid
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-megaraid"
hostgroups: megaraid
- ###
- #-
- # 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"
hostgroups: mptraid
-
- ###
-# -
-# name: RAID - megactl
-# servicegroups: raid
-# nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-megactl"
-# hostgroups: megactl
- ###
+ ###
-
name: RAID - DRBD
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/dsa-check-drbd -d All"
hostgroups: drbd-hosts
- ###
+ # }}}
+ # }}}
+ # {{{ ### mail stuff
+ # {{{ exim processes and mailq
-
- name: process - slapd
- 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: draghi
+ name: process - exim
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
+ hostgroups: computers
+ excludehostgroups: postfix-hosts, mail-relay
+ excludehosts: master, busoni, quantz, buxtehude
-
- name: network service - ldaps cert
- check: dsa_check_cert!636
- depends: process - slapd
- normal_check_interval: 60
- hosts: draghi
- ###
+ name: process - exim
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
+ hostgroups: mail-relay
-
- name: process - ulogd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C ulogd -a '/usr/sbin/ulogd -d'"
+ name: process - exim - total
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1: -C exim4"
hostgroups: computers
- excludehostgroups: freebsd, sparc
+ excludehostgroups: postfix-hosts
+ excludehosts: master, busoni, quantz, buxtehude
-
- name: unexpected process - ulogd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C ulogd"
- hostgroups: freebsd, sparc
+ name: process - exim
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:300 -c 1:500 -C exim4 -a '/usr/sbin/exim4'"
+ hosts: master, busoni, quantz, buxtehude
+ -
+ name: mail queue
+ nrpe: "/usr/lib/nagios/plugins/check_mailq -M exim -w 1000 -c 2000"
+ hostgroups: heavy-exim
+ # }}}
+ # {{{ clamav
+ -
+ name: process - clamav - clamd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u clamav -C clamd -a '/usr/sbin/clamd'"
+ hostgroups: heavy-exim, heavy-postfix
+ -
+ name: service - clamav
+ nrpe: "/usr/lib/nagios/plugins/check_clamd -H /var/run/clamav/clamd.ctl"
+ hostgroups: heavy-exim, heavy-postfix
+ depends: process - clamav - clamd
+ -
+ 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: unwanted process - clamav
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C clamd"
+ hostgroups: computers
+ excludehostgroups: heavy-exim, heavy-postfix
+ -
+ name: unwanted process - freshclam
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C freshclam"
+ hostgroups: computers
+ excludehostgroups: heavy-exim, heavy-postfix
+ # }}}
+ # {{{ anti-spam
+ -
+ name: process - spamd - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd --create-prefs --max-children 5 --helper-home-dir -d --pidfile=/var/run/spamd.pid'"
+ hostgroups: spamd
+ excludehosts: picconi
+ -
+ name: process - spamd - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd --create-prefs --max-children 20 --min-spare=5 --helper-home-dir -d --pidfile=/var/run/spamd.pid'"
+ hosts: picconi
+ -
+ name: process - spamd - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd --create-prefs --max-children 10 --helper-home-dir -d --pidfile=/var/run/spamd.pid'"
+ hosts: bendel
+ -
+ name: process - spamd - child
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:11 -c 1: -C spamd -a 'spamd child'"
+ hosts: bendel
+ hostgroups: spamd
+ depends: process - spamd - master
+ #
+ -
+ name: process - spamd - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u debbugs -C spamd -a '/usr/sbin/spamd -d '"
+ hosts: buxtehude
+ -
+ name: process - spamd - child
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1: -u debbugs -C spamd -a 'spamd child'"
+ hosts: buxtehude
+ #
+ -
+ name: unwanted process - spamd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd"
+ hostgroups: computers
+ excludehostgroups: spamd
+ excludehosts: bendel, busoni, buxtehude
- ###
-
- name: process - udevd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -p 1 -C udevd -a 'udevd'"
+ name: unwanted process - greylistd
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C greylistd"
hostgroups: computers
- excludehostgroups: freebsd
+
+ ###
-
- name: unexpected process - udev
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C udevd"
- hostgroups: freebsd
- ###
+ name: process - postgrey
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --unix=/var/run/postgrey/socket --retry-window=4 --auto-whitelist-clients=10 --exim'"
+ hostgroups: heavy-exim
-
- name: process - acpid
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C acpid -a '/usr/sbin/acpid'"
- hostgroups: acpid-hosts
+ name: process - postgrey
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --inet=127.0.0.1:60000'"
+ hostgroups: heavy-postfix
+ #
-
- name: unexpected process - acpid
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C acpid"
+ name: unwanted process - postgrey
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C postgrey"
hostgroups: computers
- excludehostgroups: acpid-hosts, kvmdomains
- ###
-# -
-# name: process - bosserver
-# nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C bosserver -a '/usr/sbin/bosserver'"
-# hostgroups: bosserver
-#
- ###
+ excludehostgroups: heavy-postfix, heavy-exim
+ ###
+ -
+ name: process - amavis - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u amavis -a 'amavisd-new (master)'"
+ hostgroups: amavis-hosts
+ -
+ name: process - amavis - all
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1:10 -u amavis -a 'amavisd-new '"
+ hostgroups: amavis-hosts
+ depends: process - amavis - master
+ #
+ -
+ name: unwanted process - amavis
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C amavisd-new"
+ hostgroups: computers
+ excludehostgroups: amavis-hosts
+ ###
+ -
+ name: process - weightd - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (master)'"
+ hostgroups: heavy-postfix, alioth
+ -
+ name: process - weightd - cache
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (cache)'"
+ hostgroups: heavy-postfix, alioth
+ depends: process - weightd - master
+ -
+ name: process - weightd - child
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:50 -c 1: -u polw -a 'policyd-weight (child)'"
+ hostgroups: heavy-postfix, alioth
+ depends: process - weightd - master
+ ###
+ -
+ name: unwanted process - policyd-weight
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C policyd-weight"
+ hostgroups: computers
+ excludehostgroups: heavy-postfix, alioth
+ # }}}
+ # {{{ postfix
+ ###
+ -
+ name: process - postfix - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C master -a '/usr/lib/postfix/master'"
+ hostgroups: postfix-hosts
+ -
+ name: process - postfix - qmgr
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C qmgr -a 'qmgr -l -t fifo -u'"
+ hostgroups: postfix-hosts
+ depends: process - postfix - master
+ -
+ name: process - postfix - pickup
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C pickup -a 'pickup -l -t fifo -u -c'"
+ hostgroups: postfix-hosts
+ depends: process - postfix - master
-
- name: process - xinetd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xinetd -a '/usr/sbin/xinetd -pidfile /var/run/xinetd.pid -stayalive'"
- hostgroups: xinetd-hosts
+ name: process - postfix - anvil
+ 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
+
-
- name: unwanted process - xinetd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C xinetd"
- hostgroups: computers
- excludehostgroups: xinetd-hosts, deadslow
- ###
+ name: process - postfix - trivial-rewrite
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C trivial-rewrite -a 'trivial-rewrite -n rewrite -t unix -u -c'"
+ hostgroups: postfix-hosts
+ depends: process - postfix - master
-
- name: network service - finger
- check: check_tcp!79
- hosts: draghi
- depends: process - xinetd
- ###
+ name: process - postfix - proxymap
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C proxymap -a 'proxymap -t unix -u'"
+ hostgroups: postfix-hosts
+ depends: process - postfix - master
-
- name: network service - rsync
- check: check_tcp!873
- hostgroups: rsyncd-hosts
- depends: process - xinetd
- excludehosts: rietz
+ name: process - postfix - cleanup
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u postfix -C cleanup -a 'cleanup -z -t unix -u -c'"
+ hostgroups: postfix-hosts
+ depends: process - postfix - master
-
- name: network service - rsync
- check: check_tcp!873
- hosts: rietz2
- depends: rietz:process - xinetd
+ name: process - postfix - local
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u postfix -C local -a 'local -t unix'"
+ hostgroups: postfix-hosts
+ depends: process - postfix - master
+
-
- name: network service - rsync
- check: check_tcp!873
- hosts: milanollo2
- depends: milanollo:process - xinetd
+ name: network service local - smtps cert
+ 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: 120
+ # }}}
+ # {{{ mail - network service
+ -
+ name: network service - smtp
+ check: dsa_check_smtp
+ hostgroups: computers
+ excludehostgroups: postfix-hosts, incomingmailrelayed, incomingmailrelayed2025
+ depends: process - exim
- ###
-
- name: process - icinga
- # there is always one extra process per check currently running..
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:30 -c 1: -u nagios -C icinga -a '/usr/sbin/icinga -d /etc/icinga/icinga.cfg'"
- hosts: tchaikovsky
- ###
+ name: network service - smtp
+ check: dsa_check_smtp
+ hostgroups: postfix-hosts
+ depends: process - postfix - master
-
- name: process - jetty - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -a 'jsvc.exec'"
- hostgroups: jetty-hosts
+ name: network service - submission
+ check: dsa_check_smtp_port!587
+ hostgroups: incomingmailrelayed
+ depends: process - exim
-
- name: process - jetty - worker
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1:100 -u jetty -a 'jsvc.exec -user jetty'"
- hostgroups: jetty-hosts
- depends: process - jetty - master
-
- ###
+ name: network service - smtp 2025
+ check: dsa_check_smtp_port!2025
+ hostgroups: incomingmailrelayed2025
+ depends: process - exim
+ # }}}
+ # }}}
+ # {{{ ### host specific services
+ # {{{ HTTP
-
name: process - apache2 - master
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -a /usr/sbin/apache2"
name: unwanted process - apache2
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C apache2"
hostgroups: computers
- excludehostgroups: apache2-hosts, deadslow
+ excludehostgroups: apache2-hosts
-
name: network service - http
check: check_https
hostgroups: apache-https
excludehosts: handel,menotti
+ excludehostgroups: broken_https_default_vhost
depends: "process - apache2 - master"
normal_check_interval: 120
-
hosts: handel,menotti
depends: "process - apache2 - master"
normal_check_interval: 120
+ -
+ name: network service - https
+ check: dsa_check_https_any_status
+ hostgroups: broken_https_default_vhost
+ depends: "process - apache2 - master"
+ normal_check_interval: 120
-
name: network service - https cert
check: dsa_check_cert!443
name: unwanted network service - https
check: dsa_check_port_closed!443
hostgroups: apache2-hosts
- excludehostgroups: apache-https, deadslow
+ excludehostgroups: apache-https
# ravel does ssh on port 443
excludehosts: ravel
normal_check_interval: 60
-
-###
- -
- name: process - varnish - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C varnishd -a '/usr/sbin/varnishd -P /var/run/varnishd.pid -a :80 -T localhost:6082 -f /etc/varnish/default.vcl'"
- hostgroups: varnish-hosts
- -
- name: process - varnish - worker
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u nobody -C varnishd -a '/usr/sbin/varnishd -P /var/run/varnishd.pid -a :80 -T localhost:6082 -f /etc/varnish/default.vcl'"
- hostgroups: varnish-hosts
- depends: process - varnish - master
- -
- name: network service - http
- check: check_http
- hostgroups: varnish-hosts
- depends: process - varnish - master
-
-###
-
- ####
- -
- name: process - named
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:15 -c 1: -u bind -C named -a '/usr/sbin/named -u bind'"
- hostgroups: bind9-hosts
- -
- name: network service - dns
- check: check_dns
- hostgroups: bind9-hosts
- depends: process - named
- -
- name: unwanted process - named
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C named"
- hostgroups: computers
- excludehostgroups: bind9-hosts, deadslow
-
- ####
+ # }}}
+ # {{{ FTP
-
name: network service - ftp
check: check_ftp
check: check_ftp
hosts: klecker-ftp
depends: klecker:process - xinetd
-
- ####
- #-
- # name: process - debianqueued
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u dak -C debianqueued"
- -
- name: process - debianqueued
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u dak-unpriv -C debianqueued"
- hostgroups: uploadqueue
-
-
- ###
- #-
- # name: process - postgresql81 - master
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postmaster -a '/usr/lib/postgresql/8.1/bin/postmaster -D /var/lib/postgresql/8.1/main -c config_file=/etc/postgresql/8.1/main/postgresql.conf'"
- # hostgroups: postgres81-hosts
- #-
- # name: process - postgresql81 - writer
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postmaster -a 'postgres: writer process'"
- # hostgroups: postgres81-hosts
- # depends: process - postgresql81 - master
- #-
- # name: process - postgresql81 - buffer
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postmaster -a 'postgres: stats buffer process'"
- # hostgroups: postgres81-hosts
- # depends: process - postgresql81 - master
- #-
- # name: process - postgresql81 - collector
- # 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 - postgresql81 - master
- ####
+ # }}}
+ # {{{ postgres
-
name: unwanted process - postgresql
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C postgres"
hostgroups: computers
- excludehostgroups: postgres91-hosts, deadslow
+ excludehostgroups: postgres91-hosts
-
name: unwanted process - postgresql 9.0
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C postgres -a '9.0/bin/postgres'"
hostgroups: computers
-
name: process - postgresql91 - master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:4 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/9.1/bin/postgres'"
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/9.1/bin/postgres'"
hostgroups: postgres91-hosts
-
name: postgresql backups
nrpe: "/usr/bin/sudo -u debbackup /usr/lib/nagios/plugins/dsa-check-backuppg"
- #hosts: beethoven
hosts: backuphost
- ####
- ####
- -
- name: process - stunnel4 - puppet-ekeyd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1: -u stunnel4 -C stunnel4 -a '/usr/bin/stunnel4 /etc/stunnel/puppet-ekeyd.conf'"
- hostgroups: squeeze, wheezy
- excludehostgroups: freebsd
- ####
- -
- name: process - UPS - nut usbhid-ups - ups1
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nut -C usbhid-ups -a '/lib/nut/usbhid-ups -a ups1'"
- hosts: franck
- -
- name: process - UPS - nut upsd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nut -C upsd -a '/sbin/upsd'"
- hosts: franck
- -
- name: process - UPS - nut upsmon master
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C upsmon -a '/sbin/upsmon'"
- hosts: franck
- -
- name: process - UPS - nut upsmon worker
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nut -C upsmon -a '/sbin/upsmon'"
- hosts: franck
- depends: process - UPS - nut upsmon master
- -
- name: UPS - on line power
- nrpe: "/usr/lib/nagios/plugins/dsa-check-ups"
- hosts: franck
- depends: process - UPS - nut upsd
- ###
+ # }}}
+ # {{{ power
+# -
+# name: process - UPS - nut usbhid-ups - ups1
+# nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nut -C usbhid-ups -a '/lib/nut/usbhid-ups -a ups1'"
+# hosts: franck
+# -
+# name: process - UPS - nut upsd
+# nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nut -C upsd -a '/sbin/upsd'"
+# hosts: franck
+# -
+# name: process - UPS - nut upsmon master
+# nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C upsmon -a '/sbin/upsmon'"
+# hosts: franck
+# -
+# name: process - UPS - nut upsmon worker
+# nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nut -C upsmon -a '/sbin/upsmon'"
+# hosts: franck
+# depends: process - UPS - nut upsmon master
+# -
+# name: UPS - on line power
+# nrpe: "/usr/lib/nagios/plugins/dsa-check-ups"
+# hosts: franck
+# depends: process - UPS - nut upsd
+ # }}}
+ # {{{ buildd
-
name: process - buildd
servicegroups: buildd
contact_groups: buildd
-
name: processes - zombie schroot
- #nrpe: "/usr/lib/nagios/plugins/check_procs -a schroot -s Zs -c 0"
nrpe: "(/usr/lib/nagios/plugins/check_procs -a schroot -s Zs -c 0 > /dev/null || /usr/lib/nagios/plugins/check_procs -a schroot -s Zs -c 0) && /usr/lib/nagios/plugins/check_procs -a schroot -s ZNs -c 0"
hostgroups: buildd
contact_groups: +buildd
nrpe: "/usr/lib/nagios/plugins/check_procs -m 'ELAPSED' -c 500 -C lvcreate -u root -a 'lvcreate'"
hostgroups: buildd
contact_groups: +buildd
- ####
- -
- name: network service - gobby
- check: check_tcp!6523
- hosts: gombert
- contact_groups: gobby
- ####
- #-
- # name: process - tftpd
- # nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C in.tftpd -a '/usr/sbin/in.tftpd -l -B 1450 -s /var/lib/tftpboot'"
- # hostgroups: tftpd-hosts
- ###
-
- ############ NFS Stuff ############
- ####
+ # }}}
+ # {{{ NFS Stuff
-
name: process - statd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u statd -C rpc.statd -a '/sbin/rpc.statd'"
name: process - mountd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C rpc.mountd -a '/sbin/rpc.mountd'"
hostgroups: nfs-server
- -
- name: nfs server glinka reachable
- nrpe: "/usr/lib/nagios/plugins/check_ping -H 192.168.2.76 -w 50,10% -c 200,30%"
- hosts: quantz
#
-
name: process - automount
name: unwanted process - automount
nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C automount"
hostgroups: computers
- excludehostgroups: autofs, deadslow
-
- ############ MISC OTHER Stuff ############
- #####
- -
- name: puppetmaster cert
- nrpe: "/usr/lib/nagios/plugins/dsa-check-cert-expire /var/lib/puppet/ssl/certs/ca.pem"
- hosts: handel
- normal_check_interval: 60
- max_check_attempts: 2
- retry_check_interval: 5
+ excludehostgroups: autofs
+ # }}}
+ # {{{ mirroring
-
name: mirror sync - bugs
check: "dsa_check_mirrorsync_skew!bugs.debian.org!project/trace/bugs-master.debian.org!120:600"
normal_check_interval: 15
max_check_attempts: 5
retry_check_interval: 5
+ # }}}
+ # {{{ DNS
+ -
+ name: process - named
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:15 -c 1: -u bind -C named -a '/usr/sbin/named -u bind'"
+ hostgroups: bind9-hosts
+ -
+ name: network service - dns
+ check: check_dns
+ hostgroups: bind9-hosts
+ depends: process - named
+ -
+ name: unwanted process - named
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C named"
+ hostgroups: computers
+ excludehostgroups: bind9-hosts
+ ###
-
name: DNS SOA sync - debian.org
check: "dsa_check_soas_add!denis.debian.org!debian.org"
name: DNS SOA sync - debian.com
check: "dsa_check_soas_add!denis.debian.org!debian.com"
hosts: global
- -
- name: DNS SOA sync - mirror.debian.net
- check: "dsa_check_soas_add!denis.debian.org!mirror.debian.net"
- hosts: global
-
name: DNS SOA sync - 144-28.118.59.86.in-addr.arpa
check: "dsa_check_soas_add!denis.debian.org!144-28.118.59.86.in-addr.arpa"
hosts: global
-
name: DNS SOA sync - alioth.debian.org
- check: "dsa_check_soas_add!alioth.debian.org!alioth.debian.org"
+ check: "dsa_check_soas_add!denis.debian.org!alioth.debian.org"
hosts: global
-
name: DNS - delegation and signature expiry
hosts: global
remotecheck: "/usr/lib/nagios/plugins/dsa-check-statusfile /srv/dns.debian.org/var/nagios/ds"
runfrom: denis
-
- ############
+ # }}}
+ # {{{ storage
-
name: ping alive check
remotecheck: "/usr/lib/nagios/plugins/check_ping -H $HOSTADDRESS$ -w 50,10% -c 200,30%"
hosts: giustini
-
name: event log
- remotecheck: "/usr/lib/nagios/plugins/dsa-check-msa-eventlog --start=7778 $HOSTADDRESS$ public"
+ remotecheck: "/usr/lib/nagios/plugins/dsa-check-msa-eventlog --start=9966 $HOSTADDRESS$ public"
runfrom: dijkstra
hosts: giustini
+ # }}}
+ # {{{ porterbox
+ -
+ name: current chroots
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-dchroots-current"
+ hostgroups: porterbox
+ normal_check_interval: 60
+ retry_check_interval: 15
+ # }}}
+ # {{{ openstack
+ -
+ name: process - openstack - memcached
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nobody -C memcached -a '/usr/bin/memcached -m 64 -p 11211 -u nobody -l 127.0.0.1'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - glance-registry
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u glance -C glance-registry -a '/usr/bin/python /usr/bin/glance-registry'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - nova-api
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nova -C nova-api -a '/usr/bin/python /usr/bin/nova-api --config-file=/etc/nova/nova.conf --log-file /var/log/nova/nova-api.log'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - nova-compute
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nova -C nova-compute -a '/usr/bin/python /usr/bin/nova-compute --config-file=/etc/nova/nova.conf --log-file /var/log/nova/nova-compute.log --config-file=/etc/nova/nova-compute.conf'"
+ hostgroups: openstack-compute
+ -
+ name: process - openstack - nova-cert
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nova -C nova-cert -a '/usr/bin/python /usr/bin/nova-cert --config-file=/etc/nova/nova.conf --log-file /var/log/nova/nova-cert.log'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - nova-conductor
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nova -C nova-conductor -a '/usr/bin/python /usr/bin/nova-conductor --config-file=/etc/nova/nova.conf --log-file /var/log/nova/nova-conductor.log'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - nova-consoleauth
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nova -a '/usr/bin/python /usr/bin/nova-consoleauth --config-file=/etc/nova/nova.conf --log-file /var/log/nova/nova-consoleauth.log'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - nova-scheduler
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nova -a '/usr/bin/python /usr/bin/nova-scheduler --config-file=/etc/nova/nova.conf --log-file /var/log/nova/nova-scheduler.log'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - nova-spicehtml5proxy
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nova -a '/usr/bin/python /usr/bin/nova-spicehtml5proxy --log-file /var/log/nova/nova-consoleproxy.log'"
+ hostgroups: openstack-controller
+ -
+ name: process - openstack - neutron-server
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u neutron -C neutron-server -a '/usr/bin/python2.7 /usr/bin/neutron-server --config-file=/etc/neutron/neutron.conf --config-file /etc/neutron/plugins/openvswitch/ovs_neutron_plugin.ini --log-file=/var/log/neutron/neutron-server.log'"
+ hostgroups: openstack-controller
+ # }}}
+ # {{{ misc
+ -
+ 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'"
+ hostgroups: dl385
+ ###
+ -
+ name: process - slapd
+ 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: draghi
+ -
+ name: network service - ldaps cert
+ check: dsa_check_cert!636
+ depends: process - slapd
+ normal_check_interval: 60
+ hosts: draghi
+ ###
+ -
+ name: network service - finger
+ check: check_tcp!79
+ hosts: draghi
+ depends: process - xinetd
+ ###
+ -
+ name: network service - rsync
+ check: check_tcp!873
+ hostgroups: rsyncd-hosts
+ depends: process - xinetd
+ excludehosts: rietz
+ -
+ name: network service - rsync
+ check: check_tcp!873
+ hosts: rietz2
+ depends: rietz:process - xinetd
+ -
+ name: network service - rsync
+ check: check_tcp!873
+ hosts: milanollo2
+ depends: milanollo:process - xinetd
+ ###
+ -
+ name: process - icinga
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:30 -c 1: -u nagios -C icinga -a '/usr/sbin/icinga -d /etc/icinga/icinga.cfg'"
+ hosts: tchaikovsky
+ ###
+ -
+ name: process - jetty - master
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -a 'jsvc.exec'"
+ hostgroups: jetty-hosts
+ -
+ name: process - jetty - worker
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1:100 -u jetty -a 'jsvc.exec -user jetty'"
+ hostgroups: jetty-hosts
+ depends: process - jetty - master
+ ###
+ -
+ name: process - debianqueued
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u dak-unpriv -C debianqueued"
+ hostgroups: uploadqueue
+ ####
+ -
+ name: network service - gobby
+ check: check_tcp!6523
+ hosts: gombert
+ contact_groups: gobby
+ ####
+ -
+ name: puppetmaster cert
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-cert-expire /var/lib/puppet/ssl/certs/ca.pem"
+ hosts: handel
+ normal_check_interval: 60
+ max_check_attempts: 2
+ retry_check_interval: 5
+ # }}}
+# }}}
# vim: set ts=2 sw=2 et ai si fdm=marker:
projects / mirror / dsa-nagios.git / blobdiff