X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=config%2Fnagios-master.cfg;h=e0a3fbb23116a57dd432f4098f8cafd8bcfe5400;hb=d4b04d6bb4e86596fb23fd1a412038a38e41aa92;hp=25de02c29cefe4931c708178b71dea861652d7f7;hpb=a9f9473d0921da8f828fdf0bb79f4d44278bf14d;p=mirror%2Fdsa-nagios.git

diff --git a/config/nagios-master.cfg b/config/nagios-master.cfg
index 25de02c..e0a3fbb 100644
--- a/config/nagios-master.cfg
+++ b/config/nagios-master.cfg
@@ -78,6 +78,10 @@ servers:
     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
@@ -205,35 +209,43 @@ servers:
   abel:
     address: 217.140.96.56
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, porterbox, wheezy
+    hostgroups: computers, hasbootfs, hassrvfs, porterbox, wheezy, broken_mq
   antheil:
     address: 217.140.96.60
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, broken_mq
   arnold:
     address: 217.140.96.57
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy, broken_mq
+  arm-arm-01:
+    address: 217.140.96.58
+    parents: gw-arm
+    hostgroups: hasbootfs, hassrvfs, buildd, wheezy, broken_mq
+  arm-arm-02:
+    address: 217.140.96.59
+    parents: gw-arm
+    hostgroups: hasbootfs, hassrvfs, buildd, wheezy, broken_mq
   harris:
     address: 217.140.96.66
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, porterbox
+    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
+    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
+    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
+    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
+    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd, broken_mq
   # }}}
   # {{{ gw-brown
   franck:
@@ -277,19 +289,19 @@ servers:
   bm-bl9:
     address: 5.153.231.249
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, acpid-hosts, service, wheezy, openstack-compute
+    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
+    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
+    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
+    hostgroups: computers, bm-bl, acpid-hosts, service, wheezy, openstack-compute, broken_mq
 
   milanollo:
     address: 5.153.231.2
@@ -336,7 +348,7 @@ servers:
   coccia:
     address: 5.153.231.11
     parents: ganeti-bytemark
-    hostgroups: computers, hassrvfs, kvmdomains, wheezy, autofs, nfs-client
+    hostgroups: computers, hassrvfs, kvmdomains, wheezy, autofs, nfs-client, uploadqueue, xinetd-hosts
   backuphost:
     address: 5.153.231.12
     parents: ganeti-bytemark
@@ -387,10 +399,6 @@ servers:
     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
-    parents: ganeti-bytemark
-    hostgroups: computers, service, kvmdomains, wheezy, bind9-hosts, no-bacula
   petrova:
     address: 5.153.231.25
     parents: ganeti-bytemark
@@ -398,7 +406,7 @@ servers:
   oyens:
     address: 5.153.231.26
     parents: ganeti-bytemark
-    hostgroups: computers, kvmdomains, wheezy, apache2-hosts, openstack-controller, apache-https
+    hostgroups: computers, kvmdomains, wheezy, apache2-hosts, openstack-controller, apache-https, broken_mq
   barriere:
     address: 5.153.231.27
     parents: ganeti-bytemark
@@ -406,11 +414,15 @@ servers:
   quantz:
     address: 5.153.231.28
     parents: ganeti-bytemark
-    hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, nfs-client, xinetd-hosts, heavy-exim, apache2-hosts, autofs
+    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
+    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, apache-https
   # }}}
   # {{{ gw-c3sl
   santoro:
@@ -449,27 +461,23 @@ servers:
   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
@@ -479,6 +487,12 @@ servers:
     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
@@ -507,27 +521,28 @@ servers:
   lw01:
     address: 185.17.185.177
     parents: gw-leaseweb
-    hostgroups: computers, service, acpid-hosts, wheezy, dl180
+    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
+    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
+    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
-  lw05:
-    address: 185.17.185.181
+    hostgroups: computers, service, acpid-hosts, wheezy, dl180, nfs-server
+  lw07:
+    address: 185.17.185.187
     parents: gw-leaseweb
-    hostgroups: computers,  service, acpid-hosts, wheezy, dl120, sw-raid
-  lw06:
-    address: 185.17.185.182
+    hostgroups: computers,  service, acpid-hosts, wheezy, dl180, nfs-client, autofs, hassrvfs, postgres91-hosts, apache2-hosts
+  lw08:
+    address: 185.17.185.189
     parents: gw-leaseweb
-    hostgroups: computers,  service, acpid-hosts, wheezy, dl120, sw-raid
+    hostgroups: computers,  service, acpid-hosts, wheezy, dl180, nfs-client, autofs, hassrvfs
+  #, apache2-hosts
   # }}}
   # {{{ gw-karlsruhe
   zemlinsky:
@@ -558,6 +573,14 @@ servers:
     address: 82.195.75.110
     parents: ganeti3
     hostgroups: computers, service, kvmdomains, wheezy, hassrvfs, spamd, heavy-exim, highload
+  mipsel-manda-01:
+    address: 82.195.75.72
+    parents: gw-arm
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy
+  mipsel-manda-02:
+    address: 82.195.75.74
+    parents: gw-arm
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, wheezy
   fils:
     address: 82.195.75.89
     parents: ganeti3
@@ -624,6 +647,14 @@ servers:
     address: 82.195.75.92
     parents: ganeti3
     hostgroups: computers, service, kvmdomains, wheezy
+  mipsel-manda-01:
+    address: 82.195.75.72
+    parents: gw-man-da
+    hostgroups: computers, buildd, wheezy, hassrvfs, sw-raid
+  mipsel-manda-02:
+    address: 82.195.75.74
+    parents: gw-man-da
+    hostgroups: computers, buildd, wheezy, hassrvfs, sw-raid
   # }}}
   # {{{ gw-marist
   zani:
@@ -653,6 +684,10 @@ servers:
 #    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
@@ -744,7 +779,7 @@ servers:
   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, xinetd-hosts, wheezy, autofs
+    hostgroups: computers, general, dl385, apache2-hosts, acpid-hosts, hasbootfs, nfs-client, hasorgfs, wheezy, autofs
   dijkstra:
     address: 206.12.19.218
     parents: sw-ubcece-kais
@@ -793,10 +828,6 @@ servers:
     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
@@ -1215,6 +1246,10 @@ hostgroups:
   security_mirror:
     alias: hosts that are security mirrors
     private: 1
+
+  broken_mq:
+    alias: hosts whose MQ is broken
+    private: 1
 # }}}
 # {{{ ############################# servicegroups #############################
 servicegroups:
@@ -1237,8 +1272,10 @@ servicegroups:
   security:
     alias: security
     servicegroup_members: apt, kernel, samhain
+  MQ:
+    alias: rabbitMQ stuff
 # }}}
-# {{{ ############################# # services #############################
+# {{{ ############################# services #############################
 services:
   # {{{ ### basic networking
   -
@@ -1339,7 +1376,7 @@ services:
     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, lw07
   -
     name: disk usage on /var/log
     servicegroups: diskspace
@@ -1400,6 +1437,31 @@ services:
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/bacula"
     hosts: beethoven
+  -
+    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
@@ -1701,20 +1763,22 @@ services:
   ###
   -
     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
+    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
-    excludehostgroups: alioth
+    excludehostgroups: alioth, broken_mq
   ###
   -
     name: local resolver
@@ -1859,11 +1923,10 @@ services:
     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
+#  -
+#    name: HW - edac status
+#    nrpe: "/usr/lib/nagios/plugins/dsa-check-edac"
+#    normal_check_interval: 120
     #hostgroups: computers
     #excludehosts: villa, lobos, senfl, schein
   -
@@ -1871,17 +1934,27 @@ services:
     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
@@ -2240,28 +2313,28 @@ services:
     hosts: backuphost
   # }}}
   # {{{ 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
+#  -
+#    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
   -
@@ -2377,7 +2450,7 @@ services:
     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
@@ -2416,7 +2489,7 @@ services:
     hosts: giustini
   -
     name: event log
-    remotecheck: "/usr/lib/nagios/plugins/dsa-check-msa-eventlog --start=8867 $HOSTADDRESS$ public"
+    remotecheck: "/usr/lib/nagios/plugins/dsa-check-msa-eventlog --start=9966 $HOSTADDRESS$ public"
     runfrom: dijkstra
     hosts: giustini
   # }}}