parents: gw-ubcece
hostgroups: layer3-infrastructure
gw-unicamp:
- address: 177.220.10.129
+ address: 143.106.167.113
parents: gw-ubcece
hostgroups: layer3-infrastructure
gw-utwente:
arm-arm-01:
address: 217.140.96.58
parents: gw-arm
- hostgroups: computers, hassrvfs, buildd, stretch, broken_mq
+ hostgroups: computers, hassrvfs, buildd, stretch, broken_mq, sw-raid
arm-arm-03:
address: 217.140.96.60
parents: gw-arm
- hostgroups: computers, hassrvfs, buildd, stretch, broken_mq
+ hostgroups: computers, hassrvfs, buildd, stretch, broken_mq, sw-raid
arm-arm-04:
address: 217.140.96.61
parents: gw-arm
- hostgroups: computers, hassrvfs, buildd, stretch, broken_mq
+ hostgroups: computers, hassrvfs, buildd, stretch, broken_mq, sw-raid
harris:
address: 217.140.96.66
parents: gw-arm
address: 140.211.166.196
parents: pieta
hostgroups: computers, hassrvfs, buildd, stretch
- powerpc-osuosl-01:
- address: 140.211.166.197
- parents: pieta
- hostgroups: computers, hassrvfs, buildd, jessie
- partch:
- address: 140.211.15.152
- parents: gw-osuosl
- hostgroups: computers, jessie, hassrvfs, porterbox, sw-raid
# }}}
# {{{ gw-sanger
sallinen:
parents: gw-sanger
hostgroups: computers, postgres94-hosts, service, apache2-hosts, sw-raid, jessie, rsyncd-hosts, hasvarlogfs, multipath-hosts, nfs-server, varnish-hosts
contacts: tjrc1, dave
- smetana:
- address: 193.62.202.29
- parents: gw-sanger
- hostgroups: computers, sw-raid, sparc, wheezy, no-bacula
- contacts: tjrc1, dave
# }}}
# {{{ gw-scanplus
lobos:
# }}}
# {{{ gw-unicamp
prokofiev:
- address: 177.220.10.140
+ address: 143.106.167.124
parents: gw-unicamp
hostgroups: computers, stretch, service, manyprocesses
- powerpc-unicamp-01:
- address: 177.220.10.141
- parents: prokofiev
- hostgroups: computers, hassrvfs, buildd, jessie
ppc64el-unicamp-01:
- address: 177.220.10.142
+ address: 143.106.167.121
parents: prokofiev
hostgroups: computers, hassrvfs, buildd, stretch
plummer:
- address: 177.220.10.143
+ address: 143.106.167.122
parents: prokofiev
hostgroups: computers, porterbox, hassrvfs, stretch
# }}}
armhf:
alias: armhf
private: 1
- sparc:
- alias: sparc
- private: 1
porterbox:
alias: developer accessible porter machines
alias: Dell PowerEdge 1950 hosts
private: 1
- wheezy:
- alias: Hosts running wheezy
jessie:
alias: Hosts running jessie
stretch:
servicegroups: diskspace
nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /storage/snapshot-farm-10"
hosts: lw10
+
+ -
+ name: disk usage on nfs farm 1
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-1"
+ hosts: lw07
+ -
+ name: disk usage on nfs farm 2
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-2"
+ hosts: lw07
+ -
+ name: disk usage on nfs farm 3
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-3"
+ hosts: lw07
+ -
+ name: disk usage on nfs farm 4
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-4"
+ hosts: lw07
+ -
+ name: disk usage on nfs farm 09
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-09"
+ hosts: lw07
+ -
+ name: disk usage on nfs farm 10
+ servicegroups: diskspace
+ nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-10"
+ hosts: lw07
+
-
name: disk usage on /srv/morgue.debian.org/
servicegroups: diskspace
name: free memory - percent
nrpe: "/usr/lib/nagios/plugins/dsa-check-memory -m pct"
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
- excludehostgroups: jessie, stretch
-
name: process - getty
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C agetty -a /sbin/agetty"
- hostgroups: jessie, stretch
+ hostgroups: computers
-
name: processes - zombies
-
name: process - ulogd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u ulog -C ulogd -a '/usr/sbin/ulogd --daemon --uid ulog'"
- hostgroups: jessie, stretch
- -
- name: unexpected process - ulogd
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C ulogd"
- hostgroups: sparc
+ hostgroups: computers
####
-
name: process - samhain
excludehostgroups: brokensamhain
# }}}
# {{{ logging
- -
- 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: jessie, stretch
-
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, stretch
+ hostgroups: computers
-
name: remote logging on lotti
name: system time synced
nrpe: "/usr/lib/nagios/plugins/dsa-check-timedatectl -s"
hostgroups: computers
- excludehostgroups: systemd-timesyncd, wheezy
+ excludehostgroups: systemd-timesyncd
servicegroups: time
-
name: system time synced
name: process - irqbalance
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C irqbalance -a '/usr/sbin/irqbalance'"
hostgroups: computers
- excludehosts: harris, smetana
+ excludehosts: harris
###
-
name: process - cron
hostgroups: computers
excludehostgroups: alioth
###
- -
- 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: alioth, jessie, stretch
- ###
-
name: MQ connection on rainier
servicegroups: MQ
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: 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: jessie, stretch
-
name: process - udevd
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -p 1 -C systemd-udevd -a '/lib/systemd/systemd-udevd'"
- hostgroups: jessie, stretch
+ hostgroups: computers
###
-
name: unexpected process - acpid
-
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: wheezy, jessie, stretch
+ hostgroups: computers
excludehostgroups: alioth
-
name: process - stunnel4 - puppet-ekeyd is crazy
runfrom: handel
# }}}
# {{{ HW health/raid
- -
- name: process - mdadm monitor
- servicegroups: raid
- nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C mdadm -a '/sbin/mdadm --monitor --pid-file /run/mdadm/monitor.pid --daemonise --scan'"
- hostgroups: sw-raid
- excludehostgroups: jessie, stretch
-
name: process - mdadm monitor
servicegroups: raid
nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C mdadm -a '/sbin/mdadm --monitor --scan'"
hostgroups: sw-raid
- excludehostgroups: wheezy
-
name: RAID - sw raid
servicegroups: raid
-
name: system - all services running
nrpe: "/usr/bin/sudo /usr/lib/nagios/plugins/dsa-check-systemd-services"
- hostgroups: jessie, stretch
+ hostgroups: computers
###
-
name: process - slapd