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
pkgmirror-1and1:
address: 213.165.95.4
parents: powell
- hostgroups: computers, service, kvmdomains, wheezy, apache2-hosts, no-bacula
+ hostgroups: computers, service, kvmdomains, wheezy, apache2-hosts, no-bacula, apache-https
babin:
address: 213.165.95.6
parents: powell
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
picconi:
address: 5.153.231.3
parents: gw-bytemark
- hostgroups: computers, service, kvmdomains, wheezy, apache2-hosts, nfs-client, autofs, heavy-exim, spamd
+ hostgroups: computers, service, kvmdomains, wheezy, apache2-hosts, nfs-client, autofs, heavy-exim, spamd, apache-https
senfter:
address: 5.153.231.4
parents: gw-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
- hostgroups: computers, general, kvmdomains, wheezy, nfs-client, autofs
+ hostgroups: computers, general, kvmdomains, wheezy, nfs-client, autofs, hassrvfs
ticharich:
address: 5.153.231.23
parents: ganeti-bytemark
hostgroups: computers, sw-raid, hassrvfs, wheezy
# }}}
# {{{ 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
- hostgroups: computers, buildd, hassrvfs, sw-raid, acpid-hosts, wheezy
+ #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
+ # hostgroups: computers, buildd, hassrvfs, sw-raid, acpid-hosts, wheezy
spohr:
address: 192.25.206.33
parents: gw-ftcollins
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
+ lw02:
+ address: 185.17.185.178
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl180
+ lw03:
+ address: 185.17.185.179
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl180
+ lw04:
+ address: 185.17.185.180
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl180
+ lw05:
+ address: 185.17.185.181
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl120, sw-raid
+ lw06:
+ address: 185.17.185.182
+ parents: gw-leaseweb
+ hostgroups: computers, service, acpid-hosts, wheezy, dl120, sw-raid
+ # }}}
# {{{ gw-karlsruhe
zemlinsky:
address: 129.143.160.6
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: 86.59.118.152
parents: gw-sil
hostgroups: computers, buildd, wheezy
+ eberlin:
+ address: 86.59.118.155
+ parents: gw-sil
+ hostgroups: computers, buildd, wheezy
# }}}
# {{{ gw-ubcece
sw-ubcece:
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
- paganini:
- address: 206.12.19.10
- parents: sw-ubcece-kais
- hostgroups: computers, hasbootfs, aacraid, hassrvfs, nfs-client, service, squeeze, autofs
respighi:
address: 206.12.19.11
parents: sw-ubcece-kais
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
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
address: 206.12.19.143
parents: ganeti2
hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, apache2-hosts, apache-https
- stanley:
- address: 206.12.19.145
- parents: ganeti2
- hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, apache2-hosts, no-bacula
muffat:
address: 206.12.19.146
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
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
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
# 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
security_mirror:
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%"
nrpe: "/usr/lib/nagios/plugins/check_disk 97 95 /srv/farm-snapshot/farm-misc"
hosts: sibelius
-
- name: disk usage on /var/lib/postgresql/9.1/dak
+ name: disk usage on /var/lib/postgresql/9.1
servicegroups: diskspace
- nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /var/lib/postgresql/9.1/dak"
+ nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /var/lib/postgresql/9.1"
hosts: franck
-
name: disk usage on /srv/ftp-master.debian.org
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/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
-
name: processes - zombies
nrpe: "/usr/lib/nagios/plugins/check_procs 5 10 -s Z"
hostgroups: computers
depends: process - ntpd
excludehosts: ancina
- excludehostgroups: ntpsuckers, deadslow
+ excludehostgroups: deadslow
servicegroups: time
#
-
remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
runfrom: lotti
hostgroups: computers
+ excludehostgroups: alioth
-
name: remote logging on lully
remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
runfrom: lully
hostgroups: computers
+ excludehostgroups: alioth
-
name: MQ connection on rainier
remotecheck: "/usr/lib/nagios/plugins/dsa-check-mq-connection $HOSTNAME$ ud dsa"
hostgroups: computers
normal_check_interval: 60
retry_check_interval: 15
+ excludehostgroups: alioth
-
name: MQ connection on rapoport
remotecheck: "/usr/lib/nagios/plugins/dsa-check-mq-connection $HOSTNAME$ ud dsa"
hostgroups: computers
normal_check_interval: 60
retry_check_interval: 15
+ excludehostgroups: alioth
### MAIL STUFF
###
-
-
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
+ 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
+ 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
+ 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, deadslow
+ excludehostgroups: heavy-postfix, deadslow, alioth
###
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
- -
- name: setup - ud-ldap freshness
- nrpe: "/usr/lib/nagios/plugins/dsa-check-udldap-freshness"
- hostgroups: computers
-
name: system - available entropy
nrpe: "/usr/lib/nagios/plugins/dsa-check-entropy"
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
+ excludehostgroups: alioth
###
-
name: process - uptimed
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
+ excludehostgroups: squeeze, 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'"
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
+ excludehostgroups: squeeze, alioth
-
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
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"
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
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
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
+ excludehostgroups: freebsd, alioth
####
-
name: process - UPS - nut usbhid-ups - ups1
############ 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
-
name: mirror sync - bugs
check: "dsa_check_mirrorsync_skew!bugs.debian.org!project/trace/bugs-master.debian.org!120:600"
remotecheck: "/usr/lib/nagios/plugins/dsa-check-msa-eventlog --start=7778 $HOSTADDRESS$ public"
runfrom: dijkstra
hosts: giustini
+ ############
+ -
+ name: current chroots
+ nrpe: "/usr/lib/nagios/plugins/dsa-check-dchroots-current"
+ hostgroups: porterbox
+ normal_check_interval: 60
+ retry_check_interval: 15
# vim: set ts=2 sw=2 et ai si fdm=marker:
projects / mirror / dsa-nagios.git / blobdiff