address: 195.20.242.124
parents: schumann
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
- hostgroups: computers, service, hasbootfs, bind9-hosts, kvmdomains, wheezy
soler:
address: 195.20.242.126
parents: schumann
pettersson:
address: 130.239.18.123
parents: gw-accumu
- hostgroups: computers, hasbootfs, aacraid, nfs-client, acpid-hosts, service, apache2-hosts, wheezy, autofs
+ hostgroups: computers, hasbootfs, aacraid, nfs-client, acpid-hosts, service, apache2-hosts, wheezy, autofs, sw-raid
praetorius:
address: 130.239.18.121
parents: gw-accumu
mips-aql-01:
address: 141.170.6.149
parents: gw-aql
- hostgroups: computers, buildd, wheezy
+ hostgroups: computers, buildd, jessie, nfs-client
mips-aql-02:
address: 141.170.6.150
parents: gw-aql
- hostgroups: computers, buildd, wheezy
+ hostgroups: computers, buildd, wheezy, nfs-client
+ minkus:
+ address: 141.170.6.151
+ parents: gw-aql
+ hostgroups: computers, porterbox, jessie, nfs-client
mipsel-aql-01:
address: 141.170.6.152
parents: gw-aql
arnold:
address: 217.140.96.57
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, broken_mq
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, jessie, broken_mq
arm-arm-01:
address: 217.140.96.58
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, broken_mq
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, jessie, broken_mq
arm-arm-02:
address: 217.140.96.59
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, broken_mq
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, jessie, broken_mq
+ arm-arm-03:
+ address: 217.140.96.60
+ parents: gw-arm
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, jessie, broken_mq
harris:
address: 217.140.96.66
parents: gw-arm
hoiby:
address: 217.140.96.71
parents: gw-arm
- hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd, broken_mq
+ hostgroups: computers, hasbootfs, hassrvfs, armhf, jessie, buildd, broken_mq
ia64-arm-01:
address: 217.140.96.61
parents: gw-arm
dinis:
address: 5.153.231.19
parents: ganeti-bytemark
- hostgroups: computers, general, kvmdomains, wheezy
+ hostgroups: computers, general, kvmdomains, wheezy, hassrvfs
donizetti:
address: 5.153.231.20
parents: ganeti-bytemark
oyens:
address: 5.153.231.26
parents: ganeti-bytemark
- hostgroups: computers, kvmdomains, wheezy, apache2-hosts, openstack-controller, apache-https, broken_mq
+ hostgroups: computers, kvmdomains, jessie, apache2-hosts, openstack-controller, apache-https, broken_mq
barriere:
address: 5.153.231.27
parents: ganeti-bytemark
lindsay:
address: 5.153.231.36
parents: ganeti-bytemark
- hostgroups: computers, service, kvmdomains, jessie
+ hostgroups: computers, service, kvmdomains, jessie, autofs, nfs-client
+ fede:
+ address: 5.153.231.37
+ hostgroups: computers, service, kvmdomains, jessie, hassrvfs
+ sor:
+ address: 5.153.231.38
+ parents: ganeti-bytemark
+ hostgroups: computers, service, kvmdomains, jessie, hassrvfs, apache2-hosts, autofs, nfs-client
# }}}
# {{{ gw-c3sl
santoro:
hostgroups: computers, buildd, hassrvfs, sw-raid, wheezy, sparc
# }}}
# {{{ gw-csail
- senfl:
- address: 128.31.0.51
- parents: gw-csail
- 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
loghost-grnet-01:
address: 194.177.211.200
parents: gw-grnet
- hostgroups: computers, service, kvmdomains, wheezy, hassrvfs
+ hostgroups: computers, service, kvmdomains, jessie, hassrvfs
+ geo3:
+ address: 194.177.211.201
+ parents: gw-grnet
+ hostgroups: computers, service, bind9-hosts, kvmdomains, jessie
+ cgi-grnet-01:
+ address: 194.177.211.202
+ parents: gw-grnet
+ hostgroups: computers, service, kvmdomains, jessie, hassrvfs, apache2-hosts, apache-https
+ x86-grnet-01:
+ address: 194.177.211.203
+ parents: ganeti-grnet
+ hostgroups: computers, buildd, hassrvfs, kvmdomains, jessie
# }}}
# {{{ gw-isc
schein:
ball:
address: 82.195.75.70
parents: gw-man-da
- hostgroups: computers, buildd, hasbootfs, wheezy
+ hostgroups: computers, buildd, hasbootfs, wheezy, sw-raid
# bartok TODO
czerny:
address: 82.195.75.109
stockhausen:
address: 82.195.75.108
parents: ganeti3
- hostgroups: computers, service, kvmdomains, wheezy, acpid-hosts, jetty-hosts
+ hostgroups: computers, service, kvmdomains, jessie, acpid-hosts
ganeti3:
address: 82.195.75.111
parents: gw-man-da
zani:
address: 148.100.88.22
parents: gw-marist
- hostgroups: computers, buildd, hassrvfs, wheezy, incomingmailrelayed, ping-suckers
+ hostgroups: computers, buildd, hassrvfs, jessie, incomingmailrelayed
# }}}
# {{{ gw-osuosl
busoni:
buxtehude:
address: 140.211.166.26
parents: byrd
- hostgroups: computers, service, hassrvfs, acpid-hosts, apache2-hosts, heavy-exim, postgres91-hosts, wheezy, hasvarlogfs, apache-https
+ hostgroups: computers, service, hassrvfs, acpid-hosts, apache2-hosts, heavy-exim, postgres91-hosts, wheezy, hasvarlogfs, apache-https, spamd
# malo TODO
mayer:
address: 140.211.166.78
parry:
address: 140.211.15.153
parents: gw-osuosl
- hostgroups: computers, wheezy, buildd, hassrvfs
+ hostgroups: computers, wheezy, buildd, hassrvfs, sw-raid
partch:
address: 140.211.15.152
parents: gw-osuosl
- hostgroups: computers, wheezy, hassrvfs, porterbox
+ hostgroups: computers, wheezy, hassrvfs, porterbox, sw-raid
rietz:
address: 140.211.166.43
parents: gw-osuosl
villa:
address: 212.211.132.32
parents: gw-scanplus-villa
- hostgroups: computers, service, apache2-hosts, rsyncd-hosts, dl380, hasvarfs, hasusrfs, hasorgfs, xinetd-hosts, wheezy, security_mirror, no-bacula
+ hostgroups: computers, service, apache2-hosts, rsyncd-hosts, dl360, hassrvfs, xinetd-hosts, jessie, security_mirror, acpid-hosts
# }}}
# {{{ gw-sil
eder:
eberlin:
address: 86.59.118.155
parents: gw-sil
- hostgroups: computers, buildd, wheezy
+ hostgroups: computers, buildd, wheezy, sw-raid
# }}}
# {{{ gw-ubcece
sw-ubcece:
# MSA 2000 (2012i)
giustini:
address: 192.168.2.6
- parents: dijkstra
+ parents: sw-ubcece-kais
hostgroups: notacomputer
# unless we implement runfrom for host alive checks
pingable: false
address: 206.12.19.15
parents: sw-ubcece-kais
hostgroups: computers, buildd, wheezy
- rossini:
- address: 206.12.19.19
- parents: sw-ubcece-kais
- hostgroups: computers, dl585, acpid-hosts, service, wheezy
- salieri:
- address: 206.12.19.20
- parents: sw-ubcece-kais
- hostgroups: computers, dl585, acpid-hosts, service, wheezy
traetta:
address: 206.12.19.21
parents: sw-ubcece-kais
# hostgroups: computers, service, hasbootfs, kvmdomains, squeeze, hasvicepa
lotti:
address: 206.12.19.121
- parents: sw-ubcece-kais
+ parents: ganeti2
hostgroups: computers, service, hasbootfs, kvmdomains, wheezy, hassrvfs
nono:
address: 206.12.19.123
- parents: traetta
+ parents: ganeti2
hostgroups: computers, service, kvmdomains, wheezy, heavy-exim, xinetd-hosts, apache2-hosts, apache-https, broken_https_default_vhost
reger:
address: 206.12.19.124
hostgroups: computers, service, kvmdomains, wheezy, postgres91-hosts, xinetd-hosts
glinka:
address: 206.12.19.126
- parents: traetta
+ parents: ganeti2
hostgroups: computers, service, kvmdomains, wheezy, apache2-hosts, nfs-client, autofs, xinetd-hosts
tye:
address: 206.12.19.129
hostgroups: computers, service, kvmdomains, wheezy, heavy-exim, apache2-hosts, nfs-client, autofs, hassrvfs
elgar:
address: 206.12.19.130
- parents: salieri
+ parents: ganeti2
hostgroups: computers, service, kvmdomains, wheezy
gombert:
address: 206.12.19.132
hostgroups: computers, service, kvmdomains, wheezy, apache2-hosts
jenkins:
address: 206.12.19.133
- parents: salieri
+ parents: ganeti2
hostgroups: computers, service, kvmdomains, wheezy
blavet:
address: 206.12.19.134
hostgroups: computers, service, kvmdomains, wheezy, xinetd-hosts, nfs-client, autofs
diabelli:
address: 206.12.19.136
- parents: traetta
+ parents: ganeti2
hostgroups: computers, service, hasbootfs, kvmdomains, wheezy, apache2-hosts, apache-https, broken_https_default_vhost
bizet:
address: 206.12.19.137
prokofiev:
address: 143.106.167.147
parents: gw-unicamp
- hostgroups: computers, jessie, acpid-hosts, service
+ hostgroups: computers, jessie, service
powerpc-unicamp-01:
address: 143.106.167.149
parents: prokofiev
- hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, jessie
+ ppc64el-unicamp-01:
+ address: 143.106.167.135
+ parents: prokofiev
+ hostgroups: computers, hasbootfs, hassrvfs, buildd, jessie
+ plummer:
+ address: 143.106.167.146
+ parents: prokofiev
+ hostgroups: computers, porterbox, hassrvfs, jessie
# }}}
# {{{ gw-utwente
klecker:
setoguchi:
address: 133.242.99.74
parents: gw-sakura
- hostgroups: computers, service, wheezy, no-bacula, hassrvfs, apache2-hosts, rsyncd-hosts, xinetd-hosts, security_mirror, apache-https
+ hostgroups: computers, service, wheezy, no-bacula, hassrvfs, apache2-hosts, rsyncd-hosts, xinetd-hosts, security_mirror, acpid-hosts
# }}}
# {{{ ############################# host groups #############################
computers:
alias: computers
private: 1
+ extinfo-icon_image: base/debian.png
+ extinfo-icon_image_alt: Debian GNU/Linux
+ extinfo-notes_url: https://db.debian.org/machines.cgi?host=%s
layer2-infrastructure:
alias: Layer 2 Devices
extinfo-icon_image: base/switch40.png
freebsd:
alias: freebsd
private: 1
+ extinfo-icon_image_alt: Debian GNU/kFreeBSD
armhf:
alias: armhf
private: 1
porterbox:
alias: developer accessible porter machines
- extinfo-icon_image: base/debian.png
- extinfo-icon_image_alt: Debian GNU/Linux
- extinfo-notes_url: http://db.debian.org/machines.cgi?host=%s
service:
alias: machines running services
- extinfo-icon_image: base/debian.png
- extinfo-icon_image_alt: Debian GNU/Linux
- extinfo-notes_url: http://db.debian.org/machines.cgi?host=%s
buildd:
alias: buildd systems
- extinfo-icon_image: base/debian.png
- extinfo-icon_image_alt: Debian GNU/Linux
- extinfo-notes_url: http://db.debian.org/machines.cgi?host=%s
general:
alias: general purpose developer accessible machines
- extinfo-icon_image: base/debian.png
- extinfo-icon_image_alt: Debian GNU/Linux
- extinfo-notes_url: http://db.debian.org/machines.cgi?host=%s
dl380:
alias: HP DL380 hosts
alias: Hosts with LSI Logic MegaRAID
private: 1
- squeeze:
- alias: Hosts running squeeze
wheezy:
alias: Hosts running wheezy
jessie:
apache2-hosts:
alias: hosts running apache2
private: 1
- jetty-hosts:
- alias: hosts running jetty
- private: 1
bind9-hosts:
alias: hosts running bind9
private: 1
high-RTT:
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
name: PING
check: "check_ping!350.0,20%!600.0,40%"
hostgroups: pingable
- excludehostgroups: layer3-infrastructure, high-RTT, ping-suckers
+ excludehostgroups: layer3-infrastructure, high-RTT
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%"
name: processes - total
nrpe: "/usr/lib/nagios/plugins/check_procs 620 700"
hostgroups: computers
+ excludehosts: prokofiev
+ -
+ name: processes - total
+ nrpe: "/usr/lib/nagios/plugins/check_procs 1500 1700"
+ hosts: prokofiev
-
name: swap usage - percent
nrpe: "/usr/lib/nagios/plugins/check_swap -w 20% -c 10%"
hostgroups: computers
normal_check_interval: 60
retry_check_interval: 15
+ -
+ name: upgraded libraries
+ servicegroups: security
+ nrpe: "sudo /usr/local/sbin/dsa-check-libs"
+ #nrpe: "sudo /usr/lib/nagios/plugins/dsa-check-libs"
+ hostgroups: computers
+ excludehostgroups: freebsd
+ normal_check_interval: 60
+ retry_check_interval: 15
+ notification_interval: 10080
-
name: installed firewall
nrpe: "/usr/lib/nagios/plugins/dsa-check-file -w -f /etc/ferm/ferm.conf"
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
+ excludehostgroups: freebsd, jessie
-
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: process - syslog-ng
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C syslog-ng -a '/sbin/syslog-ng -F'"
+ hostgroups: jessie
-
name: remote logging on lotti
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, alioth
+ excludehostgroups: freebsd, alioth
-
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, alioth, jessie
+ excludehostgroups: alioth, jessie
###
-
name: MQ connection on rainier
###
-
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'"
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xinetd -a '/usr/sbin/xinetd '"
hostgroups: xinetd-hosts
-
name: unwanted process - xinetd
-
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, jessie
+ hostgroups: wheezy, jessie
+ excludehostgroups: freebsd, alioth
+ -
+ name: process - stunnel4 - puppet-ekeyd is crazy
+ nrpe: "sudo /usr/lib/nagios/plugins/dsa-check-stunnel-sanity"
+ hostgroups: computers
excludehostgroups: freebsd, alioth
+ excludehosts: czerny, grnet-node01, storace
# }}}
# {{{ anti-services
-
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
+ -
+ name: "host SSL cert - debian server"
+ nrpe: "if [ -e /etc/ssl/debian/certs/thishost-server.crt ]; then /usr/lib/nagios/plugins/dsa-check-cert-expire /etc/ssl/debian/certs/thishost-server.crt; else echo 'No thishost-server.crt on this host.'; fi"
+ hostgroups: computers
+ -
+ name: "host SSL cert - debian client"
+ nrpe: "if [ -e /etc/ssl/debian/certs/thishost.crt ]; then /usr/lib/nagios/plugins/dsa-check-cert-expire /etc/ssl/debian/certs/thishost.crt; else echo 'No thishost.crt on this host.'; fi"
+ hostgroups: computers
# }}}
# {{{ HW health/raid
-
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-sw"
hostgroups: sw-raid
+ -
+ name: RAID - unexpected sw raid
+ servicegroups: raid
+ nrpe: "if [ -e /proc/mdstat ]; then echo 'Found /proc/mdstat'; exit 1; else echo 'No /proc/mdstat on this host.'; fi"
+ hostgroups: computers
+ excludehostgroups: sw-raid
###
-
name: HW - hpacucli status
# nrpe: "/usr/lib/nagios/plugins/dsa-check-edac"
# normal_check_interval: 120
#hostgroups: computers
- #excludehosts: villa, lobos, senfl, schein
+ #excludehosts: villa, lobos, 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, storace
+ excludehosts: villa, lobos, schein, storace
-
name: HW - hpasmcli status
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm --ps-no-redundant"
name: HW - hpasmcli status
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm --fan-no-redundant"
normal_check_interval: 120
- hosts: senfl, schein
+ hosts: schein
-
name: HW - hpasmcli status
nrpe: "/usr/lib/nagios/plugins/dsa-check-hpasm --fan-ignore-not-present"
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
+ excludehosts: bendel, busoni
-
name: unwanted process - greylistd
# {{{ 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'"
+ nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:15 -c 1: -u bind -C named -a '/usr/sbin/named '"
hostgroups: bind9-hosts
-
name: network service - dns
hostgroups: openstack-controller
# }}}
# {{{ misc
+# -
+# Disable this check until logind and binfmt_misc issues are fixed
+# something unknown is triggering mount of binfmt_misc
+# https://bugs.debian.org/772700
+# name: system - all services running
+# nrpe: "/usr/bin/sudo /bin/systemctl is-system-running"
+# hostgroups: jessie
+ ###
-
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'"
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"
projects / mirror / dsa-nagios.git / blobdiff