Decomission penalosa
[mirror/dsa-nagios.git] / config / nagios-master.cfg
index 38b28be..13295d9 100644 (file)
@@ -182,7 +182,7 @@ servers:
   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, postgres90-hosts, squeeze
+    hostgroups: computers, service, apache2-hosts, hassrvfs, hasbootfs, rsyncd-hosts, uploadqueue, kvmdomains, heavy-exim, xinetd-hosts, apache-https, postgres91-hosts, squeeze
   geo3:
     address: 195.20.242.125
     parents: schumann
@@ -210,47 +210,47 @@ servers:
   abel:
     address: 217.140.96.56
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, porterbox, squeeze
+    hostgroups: computers, hasbootfs, hassrvfs, porterbox, squeeze, deadslow
   alain:
     address: 217.140.96.58
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze, deadslow
   alwyn:
     address: 217.140.96.59
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze, deadslow
   antheil:
     address: 217.140.96.60
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze, deadslow
   arne:
     address: 217.140.96.61
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, squeeze
+    hostgroups: computers, hasbootfs, hassrvfs, squeeze, deadslow
   arnold:
     address: 217.140.96.57
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze
+    hostgroups: computers, hasbootfs, hassrvfs, buildd, squeeze, deadslow
   harris:
     address: 217.140.96.66
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, porterbox
+    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, porterbox, deadslow
   hartmann:
     address: 217.140.96.67
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd
+    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, deadslow
   hasse:
     address: 217.140.96.68
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd
+    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, deadslow
   henze:
     address: 217.140.96.70
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd
+    hostgroups: computers, hasbootfs, hassrvfs, wheezy, armhf, buildd, deadslow
   hoiby:
     address: 217.140.96.71
     parents: gw-arm
-    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd
+    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, buildd, deadslow
   # }}}
   # {{{ gw-brainfood
   master:
@@ -266,11 +266,11 @@ servers:
   franck:
     address: 128.148.34.3
     parents: gw-brown
-    hostgroups: computers, service, apache2-hosts, dl380, rsyncd-hosts, postgres90-hosts, spamd, heavy-exim, acpid-hosts, uploadqueue, xinetd-hosts, apache-https, hassrvfs, squeeze
+    hostgroups: computers, service, apache2-hosts, dl380, rsyncd-hosts, postgres91-hosts, spamd, heavy-exim, acpid-hosts, uploadqueue, xinetd-hosts, apache-https, hassrvfs, squeeze
   ries:
     address: 128.148.34.103
     parents: gw-brown
-    hostgroups: computers, service, dl385, acpid-hosts, xinetd-hosts, hassrvfs, squeeze, postgres90-hosts
+    hostgroups: computers, service, dl385, acpid-hosts, xinetd-hosts, hassrvfs, squeeze, postgres91-hosts
   # }}}
   # {{{ gw-c3sl
   santoro:
@@ -340,20 +340,6 @@ servers:
     address: 192.25.206.62
     parents: gw-ftcollins
     hostgroups: computers, buildd, hassrvfs, sw-raid, acpid-hosts, squeeze
-  penalosa:
-    address: 192.25.206.68
-    parents: gw-ftcollins
-    hostgroups: computers, buildd, hasbootfs, lenny
-    contacts: dannf
-  paer:
-    address: 192.25.206.11
-    parents: gw-ftcollins
-    hostgroups: computers, porterbox, hasbootfs, hashomefs, lenny
-  peri:
-    address: 192.25.206.15
-    parents: gw-ftcollins
-    hostgroups: computers, buildd, sw-raid, hasbootfs, lenny
-    contacts: dannf
   samosa:
     address: 192.25.206.57
     parents: gw-ftcollins
@@ -460,6 +446,10 @@ servers:
     address: 82.195.75.109
     parents: gw-man-da
     hostgroups: computers, service, dl380, acpid-hosts, squeeze
+  clementi:
+    address: 82.195.75.103
+    parents: gw-man-da
+    hostgroups: computers, service, dl380, acpid-hosts, squeeze
   bendel:
     address: 82.195.75.100
     parents: czerny
@@ -476,10 +466,6 @@ servers:
     address: 82.195.75.98
     parents: czerny
     hostgroups: computers, service, hasbootfs, kvmdomains, squeeze, hassrvfs, apache2-hosts, heavy-exim
-  kokkonen:
-    address: 82.195.75.97
-    parents: gw-man-da
-    hostgroups: computers, service, acpid-hosts, heavy-exim, mail-relay, apache2-hosts, squeeze
   lully:
     address: 82.195.75.99
     parents: czerny
@@ -535,7 +521,7 @@ servers:
   buxtehude:
     address: 140.211.166.26
     parents: byrd
-    hostgroups: computers, service, hassrvfs, acpid-hosts, apache2-hosts, heavy-exim, squeeze, hasvarlogfs
+    hostgroups: computers, service, hassrvfs, acpid-hosts, apache2-hosts, heavy-exim, postgres91-hosts, squeeze, hasvarlogfs
   # malo TODO
   mayer:
     address: 140.211.166.78
@@ -674,7 +660,7 @@ servers:
   pasquini:
     address: 206.12.19.217
     parents: sw-ubcece-kais
-    hostgroups: computers, bl460, acpid-hosts, service, squeeze
+    hostgroups: computers, bl460, acpid-hosts, service, squeeze, drbd-hosts
   luchesi:
     address: 206.12.19.214
     parents: sw-ubcece-kais
@@ -682,15 +668,15 @@ servers:
   tristano:
     address: 206.12.19.213
     parents: sw-ubcece-kais
-    hostgroups: computers, bl460, acpid-hosts, service, squeeze
+    hostgroups: computers, bl460, acpid-hosts, service, squeeze, drbd-hosts
   boito:
     address: 206.12.19.216
     parents: sw-ubcece-kais
-    hostgroups: computers, bl460, acpid-hosts, service, squeeze
+    hostgroups: computers, bl460, acpid-hosts, service, squeeze, drbd-hosts
   bertali:
     address: 206.12.19.212
     parents: sw-ubcece-kais
-    hostgroups: computers, bl460, acpid-hosts, service, squeeze
+    hostgroups: computers, bl460, acpid-hosts, service, squeeze, drbd-hosts
   ganeti2:
     address: 206.12.19.23
     parents: sw-ubcece-kais
@@ -711,10 +697,6 @@ servers:
     address: 206.12.19.113
     parents: dijkstra
     hostgroups: computers, service, hasbootfs, bind9-hosts, kvmdomains, squeeze
-  duarte:
-    address: 206.12.19.114
-    parents: dijkstra
-    hostgroups: computers, hasbootfs, hassrvfs, apache2-hosts, kvmdomains, service, squeeze, nfs-server
   morricone:
     address: 206.12.19.9
     parents: sw-ubcece-kais
@@ -766,7 +748,7 @@ servers:
   lucatelli:
     address: 206.12.19.15
     parents: sw-ubcece-kais
-    hostgroups: computers, buildd, squeeze
+    hostgroups: computers, buildd, squeeze, deadslow
   rossini:
     address: 206.12.19.19
     parents: sw-ubcece-kais
@@ -867,6 +849,14 @@ servers:
     address: 206.12.19.142
     parents: ganeti2
     hostgroups: computers, service, kvmdomains, squeeze, nfs-client, autofs, hassrvfs
+  menotti:
+    address: 206.12.19.143
+    parents: ganeti2
+    hostgroups: computers, service, kvmdomains, squeeze, hassrvfs, apache2-hosts, apache-https
+  coincy:
+    address: 206.12.19.144
+    parents: ganeti2
+    hostgroups: computers, service, kvmdomains, squeeze, hassrvfs
   # }}}
   # {{{ gw-ugent
   ancina:
@@ -911,15 +901,15 @@ servers:
   hildegard:
     address: 144.32.168.74
     parents: gw-ynic
-    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy
+    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, deadslow
   howells:
     address: 144.32.168.75
     parents: gw-ynic
-    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy
+    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, deadslow
   hummel:
     address: 144.32.168.76
     parents: gw-ynic
-    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy
+    hostgroups: computers, hasbootfs, hassrvfs, armhf, wheezy, deadslow
   porpora:
     address: 144.32.168.78
     parents: gw-ynic
@@ -963,6 +953,8 @@ hostgroups:
   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
@@ -1037,6 +1029,8 @@ hostgroups:
     alias: Hosts that are KVM domains
     private: 1
 
+  drbd-hosts:
+    alias: hosts running drbd
   postfix-hosts:
     alias: hosts running postfix instead of exim
     private: 1
@@ -1340,9 +1334,9 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_disk 95 90 /srv/farm-snapshot/farm-misc"
     hosts: sibelius
   -
-    name: disk usage on /var/lib/postgresql/9.0/dak
+    name: disk usage on /var/lib/postgresql/9.1/dak
     servicegroups: diskspace
-    nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /var/lib/postgresql/9.0/dak"
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /var/lib/postgresql/9.1/dak"
     hosts: franck
   -
     name: disk usage on /srv/ftp-master.debian.org
@@ -1354,6 +1348,26 @@ services:
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /srv/pgbackup"
     hosts: franck
+  -
+    name: disk usage on /srv/pgbackup
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/pgbackup"
+    hosts: beethoven
+  -
+    name: disk usage on /srv/git-backup
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/git-backup"
+    hosts: beethoven
+  -
+    name: disk usage on /srv/da-backup
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/da-backup"
+    hosts: beethoven
+  -
+    name: disk usage on /srv/bacula
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/bacula"
+    hosts: beethoven
 
  ############ All Computers ############
  ####
@@ -1548,7 +1562,7 @@ services:
     hostgroups: freebsd
   -
     name: process - munin-node
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u root -C munin -a '/usr/sbin/munin-node'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u root -C munin-node -a 'munin-node'"
     hostgroups: armhf
   -
     name: network service - munin-node
@@ -1568,7 +1582,7 @@ services:
     hostgroups: computers
     depends: process - ntpd
     excludehosts: ancina
-    excludehostgroups: ntpsuckers
+    excludehostgroups: ntpsuckers, deadslow
     servicegroups: time
   #
   -
@@ -1661,19 +1675,20 @@ services:
     name: unwanted process - clamav
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C clamd"
     hostgroups: computers
-    excludehostgroups: heavy-exim, heavy-postfix
+    excludehostgroups: heavy-exim, heavy-postfix, deadslow
     excludehosts: wagner
   -
     name: unwanted process - freshclam
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C freshclam"
     hostgroups: computers
-    excludehostgroups: heavy-exim, heavy-postfix
+    excludehostgroups: heavy-exim, heavy-postfix, deadslow
     excludehosts: wagner
   -
     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: powell, wagner
+    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'"
@@ -1692,17 +1707,17 @@ services:
   -
     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: busoni, buxtehude
+    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: busoni, buxtehude
+    hosts: buxtehude
   #
   -
     name: unwanted process - spamd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd"
     hostgroups: computers
-    excludehostgroups: spamd
+    excludehostgroups: spamd, deadslow
     excludehosts: bendel, busoni, wagner, buxtehude
 
  ###
@@ -1715,6 +1730,7 @@ services:
     name: unwanted process - greylistd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C greylistd"
     hostgroups: computers
+    excludehostgroups: deadslow
 
  ###
   -
@@ -1730,7 +1746,7 @@ services:
     name: unwanted process - postgrey
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C postgrey"
     hostgroups: computers
-    excludehostgroups: heavy-postfix, heavy-exim
+    excludehostgroups: heavy-postfix, heavy-exim, deadslow
  ###
   -
     name: process - amavis - master
@@ -1746,7 +1762,7 @@ services:
     name: unwanted process - amavis
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C amavisd-new"
     hostgroups: computers
-    excludehostgroups: amavis-hosts
+    excludehostgroups: amavis-hosts, deadslow
  ###
   -
     name: process - weightd - master
@@ -1767,7 +1783,7 @@ services:
     name: unwanted process - policyd-weight
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C policyd-weight"
     hostgroups: computers
-    excludehostgroups: heavy-postfix
+    excludehostgroups: heavy-postfix, deadslow
 
 
 
@@ -1889,6 +1905,7 @@ services:
     name: unwanted process - irqbalance
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance"
     hostgroups: computers
+    excludehostgroups: deadslow
 
  ####
  ###
@@ -1904,25 +1921,28 @@ services:
     name: unwanted process - gkrellmd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C gkrellmd"
     hostgroups: computers
+    excludehostgroups: deadslow
   -
     name: unwanted process - portmap
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C portmap"
     hostgroups: computers
-    excludehostgroups: nfs-client, nfs-server
+    excludehostgroups: nfs-client, nfs-server, 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
+    excludehostgroups: nfs-client, nfs-server, deadslow
   -
     name: unwanted process - inetd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C inetd"
     hostgroups: computers
     excludehosts: grieg, abel, alwyn, vasks
+    excludehostgroups: deadslow
   -
     name: unwanted process - snmpd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C snmpd"
     hostgroups: computers
+    excludehostgroups: deadslow
 
  ####
   -
@@ -2054,6 +2074,12 @@ services:
     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
  ###
   -
     name: process - slapd
@@ -2117,7 +2143,7 @@ services:
     name: unwanted process - xinetd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C xinetd"
     hostgroups: computers
-    excludehostgroups: xinetd-hosts
+    excludehostgroups: xinetd-hosts, deadslow
  ###
   -
     name: network service - finger
@@ -2173,7 +2199,7 @@ services:
     name: unwanted process - apache2
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C apache2"
     hostgroups: computers
-    excludehostgroups: apache2-hosts
+    excludehostgroups: apache2-hosts, deadslow
 
   -
     name: network service - http
@@ -2218,7 +2244,7 @@ services:
     name: unwanted network service - https
     check: dsa_check_port_closed!443
     hostgroups: apache2-hosts
-    excludehostgroups: apache-https
+    excludehostgroups: apache-https, deadslow
     # ravel does ssh on port 443
     excludehosts: ravel
     normal_check_interval: 60
@@ -2237,7 +2263,7 @@ services:
     name: unwanted process - named
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C named"
     hostgroups: computers
-    excludehostgroups: bind9-hosts
+    excludehostgroups: bind9-hosts, deadslow
 
  ####
   -
@@ -2282,7 +2308,7 @@ services:
     name: unwanted process - postresql
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C postgres"
     hostgroups: computers
-    excludehostgroups: postgres90-hosts, postgres84-hosts, postgres91-hosts
+    excludehostgroups: postgres90-hosts, postgres84-hosts, postgres91-hosts, deadslow
   -
     name: process - postresql91 - 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'"
@@ -2295,23 +2321,6 @@ services:
     name: process - postresql84 - master
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:4 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/8.4/bin/postgres'"
     hostgroups: postgres84-hosts
-  #-
-  #  name: process - postresql83 - master udd
-  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/8.3/bin/postgres -D /var/lib/postgresql/8.3/udd -c config_file=/etc/postgresql/8.3/udd/postgresql.conf'"
-  #  hosts: piatti
-  #-
-  #  name: process - postresql83 - dak master
-  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/8.3/bin/postgres -D /var/lib/postgresql/8.3/dak -c config_file=/etc/postgresql/8.3/dak/postgresql.conf'"
-  #  hosts: chopin
-  #-
-  #  name: process - postresql84 - dak master
-  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/8.4/bin/postgres -D /var/lib/postgresql/8.4/dak -c config_file=/etc/postgresql/8.4/dak/postgresql.conf'"
-  #  hosts: franck, chopin
-  #-
-  #  name: process - postresql84 - dak-dev master
-  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/8.4/bin/postgres -D /var/lib/postgresql/8.4/dak-dev -c config_file=/etc/postgresql/8.4/dak-dev/postgresql.conf'"
-  #  hosts: franck
- # backups
   -
     name: postgresql backups
     nrpe: "/usr/bin/sudo -u debbackup /usr/lib/nagios/plugins/dsa-check-backuppg"
@@ -2407,11 +2416,6 @@ services:
   #  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
  ###
-#  -
-#    name: process - dhcpd
-#    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C dhcpd3 -a '/usr/sbin/dhcpd3 -q eth0'"
-#    hosts: paer
- ###
 
  ############ NFS Stuff ############
  ####
@@ -2448,7 +2452,7 @@ services:
     name: unwanted process - automount
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C automount"
     hostgroups: computers
-    excludehostgroups: autofs
+    excludehostgroups: autofs, deadslow
 
  ############ MISC OTHER Stuff ############
  #####