X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=nagios-master.cfg;h=51ef39bfff6dd85f941cc70e471264502d862b5d;hb=224a3f5e34190c87595172a79c81516efec2ac33;hp=a10eb42c38f184363b065df46ee015b3aed67fbe;hpb=7a5d054ffe97730e782444cea8845095a5533446;p=mirror%2Fdsa-nagios.git

diff --git a/nagios-master.cfg b/nagios-master.cfg
index a10eb42..51ef39b 100644
--- a/nagios-master.cfg
+++ b/nagios-master.cfg
@@ -17,18 +17,11 @@
 #  - puccini: mailgraph
 #  -
 #  - agnesi
-#  - casals
-#  - crest
-#  - elara
-#  - europa
-#  - kassia
-#  - kullervo
 #  - lebrun
 #  - murphy
 #  - piatti
 #  - tartini
 #sarge:
-#  - caballero
 #  - spontini
 
 # down:
@@ -115,22 +108,38 @@ servers:
     address: 195.20.247.53
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+  gw-blackcat:
+    address: 193.201.200.129
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-xandros:
+    address: 142.46.212.33
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-nmmn:
+    address: 217.114.76.81
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-utwente:
+    address: 130.89.160.1
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
 
   samosa:
     address: 192.25.206.57
-    hostgroups: computers, service, dl380, apache2-hosts, bind9-hosts, heavy-email
+    hostgroups: computers, service, dl380, apache2-hosts, bind9-hosts, heavy-exim
   raff:
     address: 192.25.206.59
     parents: samosa
-    hostgroups: computers, service, dl380, apache1-hosts, bind9-hosts, rsyncd-hosts, heavy-email
+    hostgroups: computers, service, dl380, apache1-hosts, bind9-hosts, rsyncd-hosts, heavy-exim
   gluck:
     address: 192.25.206.10
     parents: samosa
-    hostgroups: computers, general, dl380, apache1-hosts, bind9-hosts, rsyncd-hosts, heavy-email
+    hostgroups: computers, general, dl380, apache1-hosts, bind9-hosts, rsyncd-hosts, heavy-exim, highload
   merkel:
     address: 192.25.206.16
     parents: samosa
-    hostgroups: computers, general, apache1-hosts, bind9-hosts, rsyncd-hosts, sw-raid, postgres81-hosts, heavy-email
+    hostgroups: computers, general, apache1-hosts, bind9-hosts, rsyncd-hosts, sw-raid, postgres81-hosts, heavy-exim
   spohr:
     address: 192.25.206.33
     parents: samosa
@@ -176,12 +185,12 @@ servers:
   master:
     address: 70.103.162.29
     parents: gw-brainfood
-    hostgroups: computers, general, apache2-hosts, bind9-hosts, heavy-email
+    hostgroups: computers, general, apache2-hosts, bind9-hosts, heavy-exim
 
   ries:
     address: 128.148.34.103
     parents: gw-brown.edu
-    hostgroups: computers, service, apache2-hosts, bind9-hosts, ftpd-hosts, dl385, rsyncd-hosts, postgres81-hosts, heavy-email
+    hostgroups: computers, service, apache2-hosts, bind9-hosts, ftpd-hosts, dl385, rsyncd-hosts, postgres81-hosts, heavy-exim
 
   mayer:
     address: 140.211.166.78
@@ -198,7 +207,7 @@ servers:
   rietz:
     address: 140.211.166.43
     parents: gw-osuosl
-    hostgroups: computers, service, apache2-hosts, bind9-hosts, rsyncd-hosts, dl385, heavy-email
+    hostgroups: computers, service, apache2-hosts, bind9-hosts, rsyncd-hosts, dl385, heavy-exim, highload
   rietz2:
     address: 140.211.166.44
     parents: rietz
@@ -217,16 +226,20 @@ servers:
     address: 128.31.0.36
     parents: gw-mit-csail
     hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, sw-raid
+  casals:
+    address: 128.31.0.16
+    parents: gw-mit-csail
+    hostgroups: deadslow
 
   klecker:
     address: 194.109.137.218
     parents: gw-xs4all
-    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, bind9-hosts, dl385, postgres81-hosts, heavy-email
+    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, bind9-hosts, dl385, postgres81-hosts, heavy-exim
 
   saens:
     address: 128.101.240.212
     parents: gw-umn.edu
-    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, dl380, heavy-email
+    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, dl380, heavy-exim
 
   argento:
     address: 195.49.152.174
@@ -273,8 +286,36 @@ servers:
 
   puccini:
     address: 87.106.4.56
-    parents: gw-frost
-    hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix
+    parents: gw-1und1
+    hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, amavis-hosts
+
+  caballero:
+    address: 193.201.200.200
+    parents: gw-blackcat
+    hostgroups: computers, buildd, sw-raid
+
+  elara:
+    address: 142.46.212.46
+    parents: gw-xandros
+    hostgroups: deadslow
+  europa:
+    address: 142.46.212.46
+    parents: gw-xandros
+    hostgroups: deadslow
+
+  kullervo:
+    address: 217.114.76.82
+    parents: gw-nmmn
+    hostgroups: deadslow
+  crest:
+    address: 217.114.76.83
+    parents: gw-nmmn
+    hostgroups: deadslow
+
+  kassia:
+    address: 130.89.175.54
+    parents: gw-utwente
+    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, dl360
 
 #############################
 # host groups
@@ -290,6 +331,8 @@ hostgroups:
     alias: Internet routers and friends
     extinfo-icon_image: base/switch40.png
     extinfo-icon_image_alt: router
+  deadslow:
+    alias: Systems too slow to run any real checks
 
   porterbox:
     alias: developer accessible porter machines
@@ -318,6 +361,9 @@ hostgroups:
   dl385:
     alias: HP DL385 hosts
     private: 1
+  dl360:
+    alias: HP DL360 hosts
+    private: 1
   sw-raid:
     alias: Hosts with Linux software raid
     private: 1
@@ -328,7 +374,7 @@ hostgroups:
   postfix-hosts:
     alias: hosts running postfix instead of exim
     private: 1
-  heavy-email:
+  heavy-exim:
     alias: "hosts running the full mail stuff, including clamav, SA, and greylistd"
     private: 1
   heavy-postfix:
@@ -343,6 +389,9 @@ hostgroups:
   bind9-hosts:
     alias: hosts running bind9
     private: 1
+  amavis-hosts:
+    alias: hosts running amavis
+    private: 1
   ftpd-hosts:
     alias: hosts running vsftpd
     private: 1
@@ -356,6 +405,10 @@ hostgroups:
     alias: hosts running postgres81
     private: 1
 
+  highload:
+    alias: "hosts on which high load is normal"
+    private: 1
+
   secondary-IPs:
     alias: secondary IP addresses
     private: 1
@@ -380,20 +433,20 @@ services:
     hostgroups: computers
   -
     name: disk usage on /
-    nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /"
+    nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /"
     hostgroups: computers
   -
     name: disk usage on /boot
-    nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /boot"
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /boot"
     hosts: sperger, rietz, steffani, penalosa, peri, albeniz, escher, goetz, mayer, mayr, paer
   -
     name: disk usage on /var
-    nrpe: "/usr/lib/nagios/plugins/check_disk 50 75 /var"
-    hosts: bartok, samosa, raff, lobos, villa, gluck, saens, escher, voltaire
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /var"
+    hosts: bartok, samosa, raff, lobos, villa, gluck, saens, escher, voltaire, puccini
   -
     name: disk usage on /org
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org"
-    hosts: bartok, sperger, samosa, raff, lobos, villa, steffani, merkel, saens, pergolesi, verdi
+    hosts: bartok, sperger, samosa, raff, lobos, villa, steffani, merkel, saens, pergolesi, verdi, puccini
   -
     name: disk usage on /srv
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /srv"
@@ -409,11 +462,11 @@ services:
   -
     name: disk usage on /tmp
     nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /tmp"
-    hosts: samosa, raff, gluck, saens, escher
+    hosts: samosa, raff, gluck, saens, escher, puccini
   -
     name: disk usage on /usr
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /usr"
-    hosts: samosa, raff, lobos, villa, gluck, saens, pergolesi
+    hosts: samosa, raff, lobos, villa, gluck, saens, pergolesi, puccini
   -
     name: disk usage on /home
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /home"
@@ -426,6 +479,10 @@ services:
     name: disk usage on /mnt/hdc
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /mnt/hdc"
     hosts: voltaire
+  -
+    name: disk usage on /x
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /x"
+    hosts: caballero
 
  ############ All Computers ############
  ####
@@ -447,6 +504,11 @@ services:
     name: load
     nrpe: "/usr/lib/nagios/plugins/check_load -w 12,10,8 -c 22,18,14"
     hostgroups: computers
+    excludehostgroups: highload
+  -
+    name: load
+    nrpe: "/usr/lib/nagios/plugins/check_load -w 100,80,60 -c 200,160,140"
+    hostgroups: highload
  ####
   -
     name: processes - zombies
@@ -484,6 +546,13 @@ services:
     depends: process - sshd
     normal_check_interval:  60
     notification_interval:  60
+
+  -
+    name: "network service - sshd"
+    check: check_ssh
+    hostgroups: deadslow
+    normal_check_interval:  180
+    notification_interval:  180
  ####
   -
     name: network service - nrpe
@@ -558,7 +627,7 @@ services:
  ###
   -
     name: process - exim
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
     hostgroups: computers
     excludehostgroups: postfix-hosts
     excludehosts: master, rietz, merkel
@@ -576,40 +645,44 @@ services:
   -
     name: process - clamav - clamd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u clamav -C clamd -a '/usr/sbin/clamd'"
-    hostgroups: heavy-email, heavy-postfix
+    hostgroups: heavy-exim, heavy-postfix
   -
     name: service - clamav
     nrpe: "/usr/lib/nagios/plugins/check_clamd -H /var/run/clamav/clamd.ctl"
-    hostgroups: heavy-email, heavy-postfix
+    hostgroups: heavy-exim, heavy-postfix
     depends: process - clamav - clamd
   -
     name: process - clamav - freshclam
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u clamav -C freshclam -a '/usr/bin/freshclam -d --quiet'"
-    hostgroups: heavy-email, heavy-postfix
+    hostgroups: heavy-exim, heavy-postfix
   #
   -
     name: unwanted process - clamav
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C clamd"
     hostgroups: computers
-    excludehostgroups: heavy-email, heavy-postfix
+    excludehostgroups: heavy-exim, heavy-postfix
   -
     name: unwanted process - freshclam
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C freshclam"
     hostgroups: computers
-    excludehostgroups: heavy-email, heavy-postfix
+    excludehostgroups: heavy-exim, heavy-postfix
  ###
   -
     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'"
-    #hosts: samosa
-    hostgroups: heavy-email
-    excludehosts: rietz
+    hostgroups: heavy-exim
+    excludehosts: rietz, merkel, raff
+  -
+    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 10 --helper-home-dir -d --pidfile=/var/run/spamd.pid'"
+    hosts: liszt
   -
     name: process - spamd - child
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1: -C spamd -a 'spamd child'"
-    hostgroups: heavy-email
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:11 -c 1: -C spamd -a 'spamd child'"
+    hosts: liszt
+    hostgroups: heavy-exim
     depends: process - spamd - master
-    excludehosts: rietz
+    excludehosts: rietz, merkel, raff
   #
   -
     name: process - spamd - master
@@ -624,19 +697,24 @@ services:
     name: unwanted process - spamd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd"
     hostgroups: computers
-    excludehostgroups: heavy-email
+    excludehostgroups: heavy-exim
+    excludehosts: liszt
+  -
+    name: unwanted process - spamd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd"
+    hosts: merkel, raff
 
  ###
   -
     name: process - greylistd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u greylist -C greylistd -a '/usr/bin/python /usr/sbin/greylistd'"
-    hostgroups: heavy-email
+    hostgroups: heavy-exim
   #
   -
     name: unwanted process - greylistd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C greylistd"
     hostgroups: computers
-    excludehostgroups: heavy-email
+    excludehostgroups: heavy-exim
 
  ###
   -
@@ -653,18 +731,18 @@ services:
   -
     name: process - amavis - master
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u amavis -C amavisd-new -a 'amavisd (master)'"
-    hostgroups: heavy-postfix
+    hostgroups: amavis-hosts
   -
     name: process - amavis - all
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1:10 -u amavis -C amavisd-new -a 'amavisd '"
-    hostgroups: heavy-postfix
+    hostgroups: amavis-hosts
     depends: process - amavis - master
   #
   -
     name: unwanted process - amavis
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C amavisd-new"
     hostgroups: computers
-    excludehostgroups: heavy-postfix
+    excludehostgroups: amavis-hosts
  ###
   -
     name: process - weightd - master
@@ -677,12 +755,18 @@ services:
     depends: process - weightd - master
   -
     name: process - weightd - child
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:30 -c 1: -u polw -C policyd-weight -a 'policyd-weight (child)'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:10 -c 1: -u polw -C policyd-weight -a 'policyd-weight (child)'"
     hostgroups: heavy-postfix
+    excludehosts: liszt
+    depends: process - weightd - master
+  -
+    name: process - weightd - child
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:40 -c 1: -u polw -C policyd-weight -a 'policyd-weight (child)'"
+    hosts: liszt
     depends: process - weightd - master
   #
   -
-    name: unwanted process - amavis
+    name: unwanted process - policyd-weight
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C policyd-weight"
     hostgroups: computers
     excludehostgroups: heavy-postfix
@@ -738,7 +822,7 @@ services:
 
   -
     name: process - postfix - smtpd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:20 -c 0:50 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'"
     hostgroups: postfix-hosts
     excludehosts: liszt
     depends: process - postfix - master
@@ -756,19 +840,19 @@ services:
     depends: process - postfix - master
   -
     name: process - postfix - smtp
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:250 -c 0:500 -u postfix -C smtp -a 'smtp -t unix -u -c'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:350 -c 0:500 -u postfix -C smtp -a 'smtp -t unix -u -c'"
     hosts: liszt
     depends: process - postfix - master
 
  ###
   -
     name: network service - smtp
-    check: check_smtp
+    check: dsa_check_smtp
     hostgroups: postfix-hosts
     depends: process - postfix - master
   -
     name: network service - smtp
-    check: check_smtp
+    check: dsa_check_smtp
     hostgroups: computers
     excludehostgroups: postfix-hosts
     depends: process - exim
@@ -793,10 +877,10 @@ services:
     name: unwanted process - system-tools-backends
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C system-tools-ba"
     hostgroups: computers
-  -
-    name: unwanted process - dbus-daemon
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C dbus-daemon"
-    hostgroups: computers
+  #-
+  #  name: unwanted process - dbus-daemon
+  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C dbus-daemon"
+  #  hostgroups: computers
   -
     name: unwanted process - gkrellmd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C gkrellmd"
@@ -813,6 +897,11 @@ services:
     name: unwanted process - inetd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C inetd"
     hostgroups: computers
+  -
+    name: unwanted process - snmpd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C snmpd"
+    hostgroups: computers
+    excludehosts: lobos, villa
 
 
  ############ Processes/Services that only run on some computers ############
@@ -842,11 +931,11 @@ services:
   -
     name: process - cpqarrayd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C cpqarrayd -a '/usr/sbin/cpqarrayd'"
-    hostgroups: dl385, dl380
+    hostgroups: dl385, dl380, dl360
   -
     name: RAID - arrayprobe
     nrpe: "sudo /usr/bin/arrayprobe"
-    hostgroups: dl385, dl380
+    hostgroups: dl385, dl380, dl360
 
  ###
   -
@@ -857,12 +946,12 @@ services:
   -
     name: process - udevd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C udevd -a 'udevd'"
-    hosts: sperger, ries, steffani, merkel, spohr, peri, penalosa, albeniz, escher, verdi, liszt
+    hosts: sperger, ries, steffani, merkel, spohr, peri, penalosa, albeniz, escher, verdi, liszt, kassia
  ###
   -
     name: process - acpid
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C acpid -a '/usr/sbin/acpid -c /etc/acpi/events -s /var/run/acpid.socket'"
-    hosts: ries, steffani, liszt
+    hosts: ries, steffani, liszt, kassia
 
  ###
   -
@@ -902,8 +991,8 @@ services:
     hosts: samosa
   -
     name: process - nagios3
-    # it forks one instance to check, so make it -w 1:2
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u nagios -C nagios3 -a '/usr/sbin/nagios3 -d /etc/nagios3/nagios.cfg'"
+    # there is always one extra process per check currently running..
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:30 -c 1: -u nagios -C nagios3 -a '/usr/sbin/nagios3 -d /etc/nagios3/nagios.cfg'"
     hosts: samosa
 
  ###
@@ -913,7 +1002,7 @@ services:
     hostgroups: apache2-hosts
   -
     name: process - apache2 - worker
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:50 -c 1:100 -u www-data -C apache2 -a /usr/sbin/apache2"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1:100 -u www-data -C apache2 -a /usr/sbin/apache2"
     hostgroups: apache2-hosts
     depends: process - apache2 - master
   -
@@ -1046,3 +1135,8 @@ services:
     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
+ ###
+  -
+    name: process - monit
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C monit -a '/usr/sbin/monit -d 30 -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
+    hosts: villa, lobos, steffani, kassia, master