X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=nagios-master.cfg;h=af3566ab8ea2435bbcfaf7234f9aeb61070d5d5c;hb=682fa1a77ef21e4b1ccd5ee3cf9a1759690d80df;hp=1a06b3db17e58dd994392fed76db78e30a11e3bf;hpb=92e95c6f5e5f6902e79cc823d75cc51081b63448;p=mirror%2Fdsa-nagios.git

diff --git a/nagios-master.cfg b/nagios-master.cfg
index 1a06b3d..af3566a 100644
--- a/nagios-master.cfg
+++ b/nagios-master.cfg
@@ -10,50 +10,27 @@
 #  - *: mailq
 #  - *: check munin stats collection works
 #  - *: check backups are successful
-#  -
-#  - agnesi
-#  - albeniz
-#  - bruckner
-#  - caballero
-#  - casals
-#  - crest
-#  - elara
-#  - escher
-#  - europa
+#  - *: unwanted: network: auth, discard, daytime, time (on some), cvs-pserver, rsync (on some), ftp (on some), http (on some)
+#  - goetz: gmond/gmetad
+#  - verdi: pg upgrade, openvpn
+#  - mundy: salinfo_decode
+#  - puccini: mailgraph
+
+# down:
+#  - sarti
 #  - goedel
-#  - goetz
-#  - kassia
-#  - kullervo
-#  - lebrun
 #  - leisner
-#  - liszt
-#  - malo
-#  - mayer
-#  - mayr
-#  - mundy
-#  - murphy
-#  - paer
-#  - pergolesi
-#  - piatti
-#  - puccini
-#  - raptopr
-#  - ravel
-#  - sarti
-#  - schulz
-#  - spontini
-#  - tartini
-#  - verdi
-#  - voltaire
 
 ---
 #############################
 # hosts
 #############################
 servers:
-  gw-MAN-DA-debian:
+  gw-man-da:
     address: 82.195.75.126
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+    contacts: joerg, bzed
   gw-HP-ftc:
     address: 192.25.206.1
     parents: samosa
@@ -94,66 +71,184 @@ servers:
     address: 195.49.152.190
     parents: gw-HP-ftc
     hostgroups: routing-infrastructure
+    contacts: bzed
+  gw-freenet:
+    address: 62.104.23.249
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-topalis:
+    address: 195.243.109.254
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-sanger:
+    address: 193.62.202.18
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+    contacts: tjrc1
+  gw-lrz:
+    address: 129.187.0.150
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-frost:
+    address: 130.81.242.195
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-1und1:
+    address: 195.20.247.54
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+    contacts: joerg
+  gw-blackcat:
+    address: 193.201.200.129
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-xandros:
+    address: 67.210.160.81
+    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
+  #gw-ughent:
+  #  address: 157.193.39.254
+  #  parents: gw-HP-ftc
+  #  hostgroups: routing-infrastructure
+  gw-agnesi:
+    address: 65.173.90.18
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-ubc:
+    address: 137.82.84.41
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+    contacts: lfilipoz
+  gw-carnet:
+    address: 161.53.160.1
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-telegraaf:
+    address: 82.94.249.153
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+  gw-helsinki:
+    address: 128.214.173.25
+    parents: gw-HP-ftc
+    hostgroups: routing-infrastructure
+    contacts: holger
 
   samosa:
     address: 192.25.206.57
-    hostgroups: computers, service, dl380, apache2-hosts, bind9-hosts, heavy-email
+    hostgroups: computers, no-udev, 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, no-udev, service, dl380, apache1-hosts, bind9-hosts, rsyncd-hosts, heavy-exim, ulogd-hosts, nfs-client
   gluck:
     address: 192.25.206.10
     parents: samosa
-    hostgroups: computers, general, dl380, apache1-hosts, bind9-hosts, rsyncd-hosts, heavy-email
+    hostgroups: computers, no-udev, general, dl380, apache1-hosts, bind9-hosts, heavy-exim, highload, nfs-client, ulogd-hosts
   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, nfs-client, ulogd-hosts
   spohr:
     address: 192.25.206.33
     parents: samosa
-    hostgroups: computers, service, dl380, apache2-hosts, postgres81-hosts
+    hostgroups: computers, service, dl380, apache2-hosts, postgres81-hosts, ulogd-hosts, nfs-server
   peri:
     address: 192.25.206.15
     parents: samosa
-    hostgroups: computers, buildd
+    hostgroups: computers, buildd, sw-raid, single-cpu
+    contacts: dannf
   penalosa:
     address: 192.25.206.68
     parents: samosa
-    hostgroups: computers, buildd
+    hostgroups: computers, buildd, sw-raid, single-cpu
+    contacts: dannf
+  mundy:
+    address: 192.25.206.62
+    parents: samosa
+    hostgroups: computers, no-udev, buildd
+  paer:
+    address: 192.25.206.11
+    parents: samosa
+    hostgroups: computers, porterbox, bind9-hosts, single-cpu
+  merulo:
+    address: 192.25.206.58
+    parents: samosa
+    hostgroups: computers, no-udev, porterbox
 
   bartok:
     address: 82.195.75.91
-    parents: gw-MAN-DA-debian
-    hostgroups: computers, service, syslog-ng-hosts, postfix-hosts, dl385
+    parents: gw-man-da
+    hostgroups: computers, service, syslog-ng-hosts, postfix-hosts, dl385, bind9-hosts
+    contacts: joerg, bzed
+  morales:
+    address: 82.195.75.97
+    parents: gw-man-da
+    hostgroups: computers, porterbox, single-cpu
+    contacts: bzed
   sperger:
     address: 82.195.75.98
-    parents: gw-MAN-DA-debian
+    parents: gw-man-da
     hostgroups: computers, porterbox, sw-raid
+    contacts: bzed
   agricola:
     address: 82.195.75.86
-    parents: gw-MAN-DA-debian
-    hostgroups: computers, porterbox, sw-raid
+    parents: gw-man-da
+    hostgroups: computers, porterbox, sw-raid, single-cpu, lenny, no-samhain
+    contacts: bzed
   arcadelt:
     address: 82.195.75.87
-    parents: gw-MAN-DA-debian
-    hostgroups: computers, buildd, sw-raid
+    parents: gw-man-da
+    hostgroups: computers, buildd, sw-raid, single-cpu, lenny, no-samhain
+    contacts: bzed
+  liszt:
+    address: 82.195.75.100
+    parents: gw-man-da
+    hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, amavis-hosts, dl385, acpid-hosts
+    contacts: bzed
+#  auber:
+#    address: 82.195.75.101
+#    parents: gw-man-da
+#    hostgroups: computers, no-udev, buildd, sw-raid
+#    contacts: bzed
 
   master:
     address: 70.103.162.29
     parents: gw-brainfood
-    hostgroups: computers, general, apache2-hosts, bind9-hosts, heavy-email
+    hostgroups: computers, no-udev, general, apache2-hosts, bind9-hosts, heavy-exim, highload
+  murphy:
+    address: 70.103.162.31
+    parents: gw-brainfood
+    hostgroups: computers, general, postfix-hosts, dl380, acpid-hosts
 
   ries:
     address: 128.148.34.103
-    parents: gw-osuosl
-    hostgroups: computers, service, apache2-hosts, bind9-hosts, ftpd-hosts, dl385, rsyncd-hosts, postgres81-hosts, heavy-email
+    parents: gw-brown.edu
+    hostgroups: computers, service, apache2-hosts, bind9-hosts, ftpd-hosts, dl385, rsyncd-hosts, postgres81-hosts, heavy-exim, acpid-hosts
 
+  mayer:
+    address: 140.211.166.78
+    parents: gw-osuosl
+    hostgroups: computers, no-udev, buildd
+  mayr:
+    address: 140.211.166.58
+    parents: gw-osuosl
+    hostgroups: computers, no-udev, buildd
+  malo:
+    address: 140.211.166.27
+    parents: gw-osuosl
+    hostgroups: computers, buildd, tftpd-hosts
   rietz:
     address: 140.211.166.43
-    parents: gw-brown.edu
-    hostgroups: computers, service, apache2-hosts, bind9-hosts, rsyncd-hosts, dl385, heavy-email
+    parents: gw-osuosl
+    hostgroups: computers, no-udev, service, apache2-hosts, bind9-hosts, rsyncd-hosts, dl385, heavy-exim, highload
   rietz2:
     address: 140.211.166.44
     parents: rietz
@@ -171,23 +266,148 @@ servers:
   steffani:
     address: 128.31.0.36
     parents: gw-mit-csail
-    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, sw-raid
+    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, sw-raid, bind9-hosts, acpid-hosts
+  mahler:
+    address: 128.31.0.46
+    parents: gw-mit-csail
+    hostgroups: computers, single-cpu, lenny, porterbox, rsyslog-hosts
 
   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, no-udev, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, dl380, heavy-exim
 
   argento:
     address: 195.49.152.174
     parents: gw-dg-i.net
+    hostgroups: computers, buildd, sw-raid, single-cpu, lenny, no-samhain
+    contacts: bzed
+
+  pergolesi:
+    address: 62.104.23.252
+    parents: gw-freenet
+    hostgroups: computers, no-udev, porterbox, sw-raid
+  bruckner:
+    address: 62.104.23.253
+    parents: gw-freenet
+    hostgroups: computers, no-udev, porterbox, single-cpu, deadslow
+
+  raptor:
+    address: 195.243.109.162
+    parents: gw-topalis
+    hostgroups: computers, porterbox, postfix-hosts
+
+  albeniz:
+    address: 193.62.202.27
+    parents: gw-sanger
+    # SMP kernel doesn't run stable
+    hostgroups: computers, porterbox, sw-raid, single-cpu
+    contacts: tjrc1
+  goetz:
+    address: 193.62.202.26
+    parents: gw-sanger
+    # SMP kernel doesn't run stable
+    hostgroups: computers, buildd, sw-raid, single-cpu
+    contacts: tjrc1
+
+  verdi:
+    address: 192.54.42.193
+    parents: gw-lrz
+    hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, postgres81-hosts, postfix-hosts
+
+  voltaire:
+    address: 72.66.115.54
+    parents: gw-frost
+    hostgroups: computers, buildd, no-samhain
+    contacts: sfrost
+
+  puccini:
+    address: 87.106.4.56
+    parents: gw-1und1
+    hostgroups: computers, no-udev, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, amavis-hosts, ulogd-hosts
+    contacts: joerg
+  powell:
+    address: 87.106.64.223
+    parents: gw-1und1
+    hostgroups: computers, service, heavy-exim, rsyncd-hosts, ulogd-hosts, acpid-hosts
+    contacts: joerg
+
+  caballero:
+    address: 193.201.200.200
+    parents: gw-blackcat
+    hostgroups: computers, no-udev, buildd, sw-raid, bind9-hosts
+
+  elara:
+    address: 67.210.160.90
+    parents: gw-xandros
+    hostgroups: deadslow
+  europa:
+    address: 67.210.160.89
+    parents: gw-xandros
+    hostgroups: deadslow
+
+  kullervo:
+    address: 217.114.76.82
+    parents: gw-nmmn
+    hostgroups: deadslow
+    contacts: luk
+  crest:
+    address: 217.114.76.83
+    parents: gw-nmmn
+    hostgroups: deadslow
+    contacts: luk
+
+  kassia:
+    address: 130.89.175.54
+    parents: gw-utwente
+    hostgroups: computers, service, postfix-hosts, apache2-hosts, ftpd-hosts, rsyncd-hosts, dl360, acpid-hosts
+
+  allegri:
+    address: 157.193.39.233
+    parents: gw-HP-ftc
+    hostgroups: computers, buildd, postfix-hosts, sw-raid, single-cpu, lenny, no-samhain
+    contacts: luk
+
+  agnesi:
+    address: 65.173.90.83
+    parents: gw-agnesi
+    hostgroups: deadslow
+
+  spontini:
+    address: 137.82.84.42
+    parents: gw-ubc
+    hostgroups: computers, buildd
+    contacts: lfilipoz
+  ravel:
+    address: 137.82.84.43
+    parents: gw-ubc
+    hostgroups: computers, general, dl385, apache2-hosts, acpid-hosts
+
+  lebrun:
+    address: 193.198.184.10
+    parents: gw-carnet
+    hostgroups: computers, buildd
+  schroeder:
+    address: 193.198.184.11
+    parents: gw-carnet
     hostgroups: computers, buildd, sw-raid
 
+  tartini:
+    address: 82.94.249.158
+    parents: gw-telegraaf
+    hostgroups: computers, sw-raid, apache2-hosts, mysql-hosts
+
+  piatti:
+    address: 193.167.161.225
+    parents: gw-helsinki
+    hostgroups: computers, postfix-hosts, dl385, ulogd-hosts, acpid-hosts
+    contacts: holger
+
 #############################
 # host groups
 #
@@ -202,6 +422,8 @@ 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
@@ -230,19 +452,39 @@ 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
+  single-cpu:
+    alias: Hosts with only one CPU
+    private: 1
+
+  lenny:
+    alias: Hosts running lenny, not etch
+    private: 1
+
+  no-samhain:
+    alias: hosts not running samhain properly
+    private: 1
 
   syslog-ng-hosts:
     alias: hosts running syslog-ng instead of sysklogd
     private: 1
+  rsyslog-hosts:
+    alias: hosts running rsyslogd instead of sysklogd
+    private: 1
   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:
+    alias: "postfix hosts running the full mail stuff, including clamav, SA, postgrey, policyd-weight"
+    private: 1
   apache2-hosts:
     alias: hosts running apache2
     private: 1
@@ -252,20 +494,72 @@ 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
+  tftpd-hosts:
+    alias: hosts running a tftpd (tftpd-hpa as a daemon)
+    private: 1
   rsyncd-hosts:
     alias: hosts providing rsync services via xinetd
     private: 1
   postgres81-hosts:
     alias: hosts running postgres81
     private: 1
+  mysql-hosts:
+    alias: hosts running mysql
+    private: 1
+  no-udev:
+    alias: hosts not using udev
+    private: 1
+  ulogd-hosts:
+    alias: hosts running ulogd
+    private: 1
+  acpid-hosts:
+    alias: hosts running acpid
+    private: 1
+
+  nfs-client:
+    alias: hosts mounting filesystems using NFS
+    private: 1
+  nfs-server:
+    alias: hosts serving filesystems using NFS
+    private: 1
+
+  highload:
+    alias: "hosts on which high load is normal"
+    private: 1
 
   secondary-IPs:
     alias: secondary IP addresses
     private: 1
 
+
+#############################
+# servicegroups
+#############################
+servicegroups:
+  diskspace:
+    alias: diskusage checks
+  buildd:
+    alias: buildd checks
+  raid:
+    alias: raid checks
+  kernel:
+    alias: kernel checks
+  weaksshkeys:
+    alias: weak ssh keys
+  apt:
+    alias: apt upgrade status
+  samhain:
+    alias: samhain integrity status
+  security:
+    alias: security
+    servicegroup_members: apt, weaksshkeys, kernel, samhain
+
 #############################
 # services
 #############################
@@ -274,6 +568,17 @@ services:
     name: PING
     check: "check_ping!300.0,20%!600.0,40%"
     hostgroups: all
+    excludehostgroups: routing-infrastructure
+    normal_check_interval: 5
+    max_check_attempts: 4
+    retry_check_interval: 1
+  -
+    name: PING
+    check: "check_ping!2000.0,60%!3000.0,80%"
+    hostgroups: routing-infrastructure
+    normal_check_interval: 5
+    max_check_attempts: 4
+    retry_check_interval: 1
 
  ############ Services ############
  ###
@@ -282,59 +587,141 @@ services:
  ####
   -
     name: disk usage - all
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 90 95"
     hostgroups: computers
   -
     name: disk usage on /
-    nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /"
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /"
     hostgroups: computers
+    excludehosts: ries
+  -
+    name: disk usage on /
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /"
+    hosts: ries
   -
     name: disk usage on /boot
-    nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /boot"
-    hosts: sperger, rietz, steffani, penalosa, peri
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /boot"
+    hosts: sperger, rietz, steffani, penalosa, peri, albeniz, goetz, mayer, mayr, paer, spontini, tartini, morales, ravel, schroeder, piatti
   -
     name: disk usage on /var
-    nrpe: "/usr/lib/nagios/plugins/check_disk 50 75 /var"
-    hosts: bartok, samosa, raff, lobos, villa, gluck, saens
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /var"
+    hosts: bartok, samosa, raff, lobos, villa, gluck, saens, voltaire, puccini, lebrun, tartini, morales, powell
   -
     name: disk usage on /org
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org"
-    hosts: bartok, sperger, samosa, raff, lobos, villa, steffani, merkel, saens
+    hosts: sperger, samosa, raff, lobos, villa, steffani, saens, pergolesi, verdi, puccini, spontini, ravel, mahler, schroeder, piatti
+  -
+    name: disk usage on /org
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /org"
+    hosts: merkel, bartok
   -
     name: disk usage on /srv
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /srv"
-    hosts: agricola, arcadelt, argento
+    hosts: agricola, arcadelt, argento, allegri, tartini, morales, powell
   -
-    name: disk usage on /org/scratch2
-    nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org/scratch2"
-    hosts: merkel
-  -
-    name: disk usage on /oldorg
-    nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /oldorg"
+    name: disk usage on /org/scratch
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org/scratch"
     hosts: merkel
   -
     name: disk usage on /tmp
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /tmp"
-    hosts: samosa, raff, gluck, saens
+    hosts: samosa, raff, gluck, saens, puccini, merkel, tartini, powell, piatti
   -
     name: disk usage on /usr
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /usr"
-    hosts: samosa, raff, lobos, villa, gluck, saens
+    hosts: samosa, raff, lobos, villa, gluck, saens, pergolesi, puccini, merulo, tartini, morales, powell
   -
     name: disk usage on /home
+    servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /home"
+    hosts: raptor, voltaire, lebrun
+  -
+    name: disk usage on /home
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /home"
     hosts: gluck
+  -
+    name: disk usage on /chroot
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /chroot"
+    hosts: raptor
+  -
+    name: disk usage on /mnt/hdc
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /mnt/hdc"
+    hosts: voltaire
+  -
+    name: disk usage on /mnt/sdb1
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /mnt/sdb1"
+    hosts: spontini
+  -
+    name: disk usage on /x
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /x"
+    hosts: caballero
 
  ############ All Computers ############
  ####
-  #-
-  #  name: apt - security updates
-  #  nrpe: "/usr/local/bin/nagios-check-apt-updates --warnifupdates"
-  #  hostgroups: computers
-  #  normal_check_interval: 480
-  #  notification_interval: 480
-  #  max_check_attempts: 4
-  #  retry_check_interval: 12
+  -
+    name: apt - security updates
+    servicegroups: apt
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-statusfile /var/cache/dsa/nagios/apt"
+    hostgroups: computers
+    normal_check_interval:  60
+    retry_check_interval: 15
+ ####
+  -
+    name: backup
+    nrpe: "sudo /usr/lib/nagios/plugins/dsa-check-dabackup"
+    hostgroups: computers
+    normal_check_interval: 180
+    max_check_attempts: 2
+    retry_check_interval: 5
+  -
+    name: backup server config
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-dabackup-server"
+    hosts: bartok
+    normal_check_interval: 180
+    max_check_attempts: 2
+    retry_check_interval: 5
+
+ ####
+  -
+    name: running kernel
+    servicegroups: kernel
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-running-kernel"
+    hostgroups: computers
+    normal_check_interval: 180
+    retry_check_interval: 5
+
+ ####
+  -
+    name: process - samhain
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C samhain -a '/usr/sbin/samhain'"
+    hostgroups: computers
+    excludehostgroups: no-samhain
+  -
+    name: samhain
+    servicegroups: samhain
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-samhain -t 45"
+    hostgroups: computers
+    excludehostgroups: no-samhain
+    depends: process - samhain
+    normal_check_interval: 240
+    retry_check_interval: 5
+
  ####
   -
     name: users
@@ -343,8 +730,18 @@ services:
  ####
   -
     name: load
-    nrpe: "/usr/lib/nagios/plugins/check_load -w 12,10,8 -c 22,18,14"
+    nrpe: "/usr/lib/nagios/plugins/check_load -w 30,28,26 -c 50,45,50"
     hostgroups: computers
+    excludehostgroups: highload
+  -
+    name: load
+    nrpe: "/usr/lib/nagios/plugins/check_load -w 140,120,100 -c 240,220,200"
+    hostgroups: highload
+    excludehosts: rietz
+  -
+    name: load
+    nrpe: "/usr/lib/nagios/plugins/check_load -w 200,200,200 -c 350,350,350"
+    hosts: rietz
  ####
   -
     name: processes - zombies
@@ -377,21 +774,74 @@ services:
     hostgroups: computers
   -
     name: "network service - sshd"
-    check: check_ssh
+    check: dsa_check_ssh
     hostgroups: computers
     depends: process - sshd
     normal_check_interval:  60
-    notification_interval:  60
+    notification_interval: 1440
+
+  -
+    name: "network service - sshd"
+    check: dsa_check_ssh
+    hostgroups: deadslow
+    excludehosts: agnesi
+    normal_check_interval:  180
+  -
+    name: "network service - sshd - 2260"
+    check: dsa_check_ssh_port!2260
+    hosts: agnesi
+    normal_check_interval:  180
+  -
+    name: "network service - sshd - 443"
+    check: dsa_check_ssh_port!443
+    hosts: gluck
+    normal_check_interval:  180
+
+  -
+    name: "network service - sshd - version"
+    check: "dsa_check_ssh_port_version!22!OpenSSH_4.3p2 Debian-9etch3"
+    depends: network service - sshd
+    hostgroups: computers, deadslow
+    excludehosts: agnesi, crest, kullervo
+    excludehostgroups: lenny
+    normal_check_interval:  360
+  -
+    name: "network service - sshd - version"
+    check: "dsa_check_ssh_port_version!22!OpenSSH_4.3p2 Debian-9etch2+m68k1"
+    depends: network service - sshd
+    hosts: crest, kullervo
+    normal_check_interval:  360
+  -
+    name: "network service - sshd - version"
+    check: "dsa_check_ssh_port_version!22!OpenSSH_5.1p1 Debian-3"
+    depends: network service - sshd
+    hostgroups: lenny
+    normal_check_interval:  360
+  -
+    name: "network service - sshd - version - 2260"
+    check: "dsa_check_ssh_port_version!2260!OpenSSH_4.3p2 Debian-9etch3"
+    depends: network service - sshd - 2260
+    hosts: agnesi
+    normal_check_interval:  360
+ #
+  -
+    name: ssh - weak keys
+    servicegroups: weaksshkeys
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-statusfile /var/cache/dsa/nagios/weak-ssh-keys"
+    hostgroups: computers
+    normal_check_interval:  360
  ####
   -
     name: network service - nrpe
     check: check_tcp!5666
     hostgroups: computers
-    max_check_attempts: -1
+    max_check_attempts: -2
+    notification_interval: 1440
   -
     name: process - nrpe
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u nagios -C nrpe -a '/usr/sbin/nrpe -c /etc/nagios/nrpe.cfg -d'"
     hostgroups: computers
+    max_check_attempts: -1
     depends: network service - nrpe
  ###
   -
@@ -410,11 +860,20 @@ services:
     # etch: nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u ntp -C ntpd -a '/usr/sbin/ntpd -p /var/run/ntpd.pid'"
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -C ntpd -a '/usr/sbin/ntpd -p /var/run/ntpd.pid'"
     hostgroups: computers
+    excludehosts: raptor
   -
     name: network service - ntp
     check: check_ntp
     hostgroups: computers
     depends: process - ntpd
+    excludehosts: raptor, allegri
+  #
+  -
+    name: network service - time
+    check: dsa_check_time
+    hosts: raptor, allegri
+    depends: process - xinetd
+
  ###
   -
     name: process - atd
@@ -423,7 +882,7 @@ services:
  ###
   -
     name: process - cron
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C cron -a /usr/sbin/cron"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u root -C cron -a /usr/sbin/cron"
     hostgroups: computers
 
  ###
@@ -431,23 +890,28 @@ services:
     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: syslog-ng-hosts
+ ###
+  -
+    name: process - rsyslogd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C rsyslogd  -a '/usr/sbin/rsyslogd -c3'"
+    hostgroups: rsyslog-hosts
  ###
   -
     name: process - syslogd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C syslogd  -a '/sbin/syslogd'"
     hostgroups: computers
-    excludehostgroups: syslog-ng-hosts
+    excludehostgroups: syslog-ng-hosts, rsyslog-hosts
   -
     name: process - klogd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C klogd  -a '/sbin/klogd -x'"
     hostgroups: computers
-    excludehostgroups: syslog-ng-hosts
+    excludehostgroups: syslog-ng-hosts, rsyslog-hosts
 
  ### MAIL STUFF
  ###
   -
     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
@@ -465,40 +929,53 @@ 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
+    hostgroups: heavy-exim, heavy-postfix
   -
     name: service - clamav
     nrpe: "/usr/lib/nagios/plugins/check_clamd -H /var/run/clamav/clamd.ctl"
-    hostgroups: heavy-email
+    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
+    hostgroups: heavy-exim, heavy-postfix
+  -
+    name: process - clamav - getsigs
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u clamav -C getclamsigs -a 'getclamsigs'"
+    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
+    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
+    excludehostgroups: heavy-exim, heavy-postfix
+  -
+    name: unwanted process - clamav - getsigs
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C getclamsigs"
+    hostgroups: computers
+    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
@@ -513,19 +990,74 @@ 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
+
+ ###
+  -
+    name: process - postgrey
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -C postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --inet=127.0.0.1:60000'"
+    hostgroups: heavy-postfix
+  #
+  -
+    name: unwanted process - postgrey
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C postgrey"
+    hostgroups: computers
+    excludehostgroups: heavy-postfix
+ ###
+  -
+    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: 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: 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: amavis-hosts
+ ###
+  -
+    name: process - weightd - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -C policyd-weight -a 'policyd-weight (master)'"
+    hostgroups: heavy-postfix
+  -
+    name: process - weightd - cache
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -C policyd-weight -a 'policyd-weight (cache)'"
+    hostgroups: heavy-postfix
+    depends: process - weightd - master
+  -
+    name: process - weightd - child
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:50 -c 1: -u polw -C policyd-weight -a 'policyd-weight (child)'"
+    hostgroups: heavy-postfix
+    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
+
 
 
  ###
@@ -550,7 +1082,7 @@ services:
     depends: process - postfix - master
   -
     name: process - postfix - anvil
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C anvil -a 'anvil -l -t unix -u'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:1 -c 0: -u postfix -C anvil -a 'anvil -l -t unix -u'"
     hostgroups: postfix-hosts
     depends: process - postfix - master
 
@@ -564,11 +1096,6 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C proxymap -a 'proxymap -t unix -u'"
     hostgroups: postfix-hosts
     depends: process - postfix - master
-  -
-    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'"
-    hostgroups: postfix-hosts
-    depends: process - postfix - master
   -
     name: process - postfix - cleanup
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u postfix -C cleanup -a 'cleanup -z -t unix -u -c'"
@@ -580,21 +1107,78 @@ services:
     hostgroups: postfix-hosts
     depends: process - postfix - master
 
-
- ###
   -
-    name: network service - smtp
-    check: check_smtp
+    name: process - postfix - smtpd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:40 -c 0:90 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'"
     hostgroups: postfix-hosts
+    excludehosts: liszt
+    depends: process - postfix - master
+  -
+    name: process - postfix - smtp
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C smtp -a 'smtp -t unix -u -c'"
+    hostgroups: postfix-hosts
+    excludehosts: liszt
+    depends: process - postfix - master
+
+  -
+    name: process - postfix - smtpd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:150 -c 0:200 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'"
+    hosts: liszt
     depends: process - postfix - master
+  -
+    name: process - postfix - smtp
+    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: computers
     excludehostgroups: postfix-hosts
     depends: process - exim
 
+  -
+    name: network service - smtp
+    check: dsa_check_smtp
+    hostgroups: postfix-hosts
+    excludehosts: verdi, kassia, allegri, raptor, piatti
+    depends: process - postfix - master
+  -
+    name: network service - smtp - port 2025
+    check: dsa_check_smtp_port!2025
+    hosts: verdi, kassia, murphy, allegri, piatti
+    depends: process - postfix - master
+  -
+    name: network service - smtp - port 8080
+    check: dsa_check_smtp_port!8080
+    hosts: murphy
+    depends: process - postfix - master
+  -
+    name: network service - smtp - port 2025
+    remotecheck: /usr/lib/nagios/plugins/check_smtp -t 40 -H $HOSTADDRESS$ -p 2025
+    runfrom: murphy
+    hosts: raptor
+    depends: process - postfix - master
+
+  -
+    name: network service local - smtps cert
+    nrpe: "/usr/lib/nagios/plugins/check_http -H localhost -p 465 -S -C 14 -t 45"
+    hostgroups: postfix-hosts
+    depends: process - postfix - master
+    normal_check_interval: 120
+
 
+  -
+    name: setup - dsa config
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-config"
+    hostgroups: computers
+    normal_check_interval: 120
+  -
+    name: setup - ud-ldap freshness
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-udldap-freshness"
+    hostgroups: computers
  ###
   -
     name: process - uptimed
@@ -605,8 +1189,45 @@ services:
     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: arcadelt, agricola, argento, penalosa, peri
+    excludehostgroups: single-cpu
+  -
+    name: unwanted process - irqbalance
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance"
+    hostgroups: single-cpu
 
+ ####
+ ###
+  #-
+  #  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 - gkrellmd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C gkrellmd"
+    hostgroups: computers
+  -
+    name: unwanted process - portmap
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C portmap"
+    hostgroups: computers
+    excludehostgroups: nfs-client, nfs-server
+  -
+    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
+  -
+    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 ############
@@ -615,7 +1236,7 @@ services:
   -
     name: process - rngd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C rngd  -a '/usr/sbin/rngd -r /dev/hwrng'"
-    hosts: bartok
+    hostgroups: dl385
  ###
   -
     name: process - sensord
@@ -625,45 +1246,105 @@ services:
  ###
   -
     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 /var/run/mdadm/monitor.pid --daemonise --scan'"
     hostgroups: sw-raid
   -
     name: RAID - sw raid
+    servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-sw"
     hostgroups: sw-raid
 
  ###
   -
     name: process - cpqarrayd
+    servicegroups: raid
     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
+    servicegroups: raid
     nrpe: "sudo /usr/bin/arrayprobe"
-    hostgroups: dl385, dl380
+    hostgroups: dl385, dl380, dl360
+  -
+    name: HW - hpacucli status
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-hpacucli"
+    normal_check_interval: 120
+    hostgroups: dl385, dl380, dl360
+ ###
+  -
+    name: RAID - areca
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-areca"
+    hosts: powell
+ ###
+  -
+    name: RAID - DAC960
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-dac960"
+    hosts: verdi
+ ###
+  -
+    name: RAID - 3ware
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-3ware"
+    hosts: puccini
+ ###
+  -
+    name: RAID - MPT
+    servicegroups: raid
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-mpt"
+    hosts: master
 
  ###
   -
     name: process - slapd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:20 -c 1:50 -u openldap -C slapd -a '/usr/sbin/slapd -g openldap -u openldap'"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:20 -c 1:50 -u openldap -C slapd -a '/usr/sbin/slapd -h ldap:/// ldaps:/// -g openldap -u openldap'"
     hosts: samosa
+ ###
+  -
+    name: process - ulogd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C ulogd -a '/usr/sbin/ulogd -d'"
+    hostgroups: ulogd-hosts
+  -
+    name: unexpected process - ulogd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C ulogd"
+    hostgroups: computers
+    excludehostgroups: ulogd-hosts
  ###
   -
     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
+    hostgroups: computers
+    excludehostgroups: no-udev
+  -
+    name: unexpected process - udev
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C udevd"
+    hostgroups: no-udev
  ###
   -
     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
+    hostgroups: acpid-hosts
+  -
+    name: unexpected process - acpid
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C acpid"
+    hostgroups: computers
+    excludehostgroups: acpid-hosts
 
  ###
   -
     name: process - xinetd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xinetd -a '/usr/sbin/xinetd -pidfile /var/run/xinetd.pid -stayalive'"
-    hosts: samosa
+    hosts: samosa, raptor, allegri, gluck
     hostgroups: rsyncd-hosts
+  -
+    name: unwanted process - xinetd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C xinetd"
+    hostgroups: computers
+    excludehosts: samosa, raptor, allegri, gluck
+    excludehostgroups: rsyncd-hosts
  ###
   -
     name: network service - finger
@@ -684,14 +1365,10 @@ services:
     depends: rietz:process - xinetd
 
  ###
-  -
-    name: process - nagios1
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u nagios -C nagios -a '/usr/sbin/nagios -d /etc/nagios/nagios.cfg'"
-    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
 
  ###
@@ -701,7 +1378,7 @@ 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
   -
@@ -736,17 +1413,15 @@ services:
   -
     name: network service - https
     check: check_https
-    hosts: samosa
+    hosts: samosa, ries
     depends: "process - apache2 - master"
     normal_check_interval: 120
-    notification_interval: 120
   -
     name: network service - https cert
     check: dsa_check_cert!443
-    hosts: samosa
+    hosts: samosa, ries
     depends: network service - https
     normal_check_interval: 240
-    notification_interval: 240
  ####
   -
     name: process - named
@@ -757,6 +1432,11 @@ services:
     check: check_dns
     hostgroups: bind9-hosts
     depends: process - named
+  -
+    name: unwanted process - named
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C named"
+    hostgroups: computers
+    excludehostgroups: bind9-hosts
 
  ####
   -
@@ -765,7 +1445,7 @@ services:
     hostgroups: ftpd-hosts
   -
     name: process - vsftp - instance
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:30 -u ftp -C vsftpd -a 'vsftpd: '"
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:80 -c 0:100 -u ftp -C vsftpd -a 'vsftpd: '"
     hostgroups: ftpd-hosts
   -
     name: network service - ftp
@@ -776,8 +1456,8 @@ services:
  ####
   -
     name: process - debianqueued
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u dak -C debianqueued -a '/usr/bin/perl -w ./debianqueued'"
-    hosts: ries
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u dak -C debianqueued"
+    hosts: ries, ravel
 
  ###
   -
@@ -799,9 +1479,33 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postmaster -a 'postgres: stats collector process'"
     hostgroups: postgres81-hosts
     depends: process - postresql81 - master
+ ####
+  -
+    name: process - mysql - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1: -u root -C mysqld_safe -a '/bin/sh /usr/bin/mysqld_safe'"
+    hostgroups: mysql-hosts
+  -
+    name: process - mysql - workers
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u mysql -C mysqld -a /usr/sbin/mysqld"
+    hostgroups: mysql-hosts
+    depends: process - mysql - master
 
  ####
-  # XXX is this needed
+  #-
+  #  name: process - xenconsoled
+  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xenconsoled -a 'xenconsoled'"
+  #  hosts: piatti
+  #-
+  #  name: process - xenstored
+  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C xenstored -a '/usr/lib/xen-3.0.3-1/bin/xenstored --pid-file /var/run/xenstore.pid'"
+  #  hosts: piatti
+  #-
+  #  name: process - xend
+  #  nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 2: -u root -C python -a 'python /usr/lib/xen-3.0.3-1/bin/xend start'"
+  #  hosts: piatti
+#
+ ####
+  # XXX is this needed?
   -
     name: process - snmpd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u snmp -C snmpd -a '/usr/sbin/snmpd -Lsd -Lf /dev/null -u snmp -I -smux -p /var/run/snmpd.pid 127.0.0.1'"
@@ -810,5 +1514,51 @@ services:
  ###
   -
     name: process - buildd
+    servicegroups: buildd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u buildd -C buildd '/usr/bin/perl /usr/bin/buildd'"
     hostgroups: buildd
+
+ ###
+  -
+    name: process - iscsid
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 1: -u root -C iscsid '/usr/sbin/iscsid'"
+    hosts: raptor
+
+ ###
+  -
+    name: process - tftpd
+    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
+ ###
+  -
+    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
+
+ ############ NFS Stuff ############
+ ####
+  -
+    name: process - portmap
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u daemon -C portmap -a '/sbin/portmap'"
+    hostgroups: nfs-client, nfs-server
+  -
+    name: process - statd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u statd -C rpc.statd -a '/sbin/rpc.statd'"
+    hostgroups: nfs-client, nfs-server
+  -
+    name: process - nfsd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u root -C nfsd -a '[nfsd]'"
+    hostgroups: nfs-server
+  -
+    name: process - lockd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C lockd -a '[lockd]'"
+    hostgroups: nfs-server
+  -
+    name: process - mountd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C rpc.mountd -a '/sbin/rpc.mountd'"
+    hostgroups: nfs-server