From c261ac73df8c3a0c6f042986da376b39dedf5763 Mon Sep 17 00:00:00 2001
From: Peter Palfrader <peter@palfrader.org>
Date: Mon, 26 May 2014 09:36:08 +0200
Subject: [PATCH] re-organize nagios master config

---
 config/nagios-master.cfg | 965 +++++++++++++++++++--------------------
 1 file changed, 471 insertions(+), 494 deletions(-)

diff --git a/config/nagios-master.cfg b/config/nagios-master.cfg
index 203efb7..c7d739a 100644
--- a/config/nagios-master.cfg
+++ b/config/nagios-master.cfg
@@ -993,14 +993,7 @@ servers:
   # }}}
   # }}}
 
-
-# {{{ hostgroups
-#############################
-# host groups
-#
-# hostgroups ircd and all are automatically defined
-#
-#############################
+# {{{ ############################# host groups #############################
 hostgroups:
   computers:
     alias: computers
@@ -1249,10 +1242,9 @@ servicegroups:
     alias: security
     servicegroup_members: apt, kernel, samhain
 # }}}
-
 # {{{ ############################# # services #############################
 services:
-  # {{{ ############ basic networking ############
+  # {{{ ### basic networking
   -
     name: PING
     check: "check_ping!350.0,20%!600.0,40%"
@@ -1283,7 +1275,7 @@ services:
     max_check_attempts: 4
     retry_check_interval: 1
   # }}}
-  # {{{ ############ Disk Usage ############
+  # {{{ ### disk usage
   -
     name: disk usage - all
     servicegroups: diskspace
@@ -1413,16 +1405,74 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/bacula"
     hosts: beethoven
   # }}}
-
-  # {{{ ############ All Computers ############
+  # {{{ ### system
+  # {{{ setup
   -
-    name: apt - security updates
-    servicegroups: apt
-    nrpe: "/usr/lib/nagios/plugins/dsa-check-statusfile /var/cache/dsa/nagios/apt"
+    name: setup - dsa config
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-config"
+    hostgroups: computers
+    normal_check_interval: 60
+    excludehostgroups: alioth
+  -
+    name: setup - local hostname etc-hosts
+    nrpe: 'if getent ahosts `hostname` | grep -q 127.0; then echo "Warning: local hostname resolves to 127/8 address"; exit 1; else echo "OK: Hostname resolves to non-127/8 address."; exit 0; fi'
+    hostgroups: computers
+    normal_check_interval: 60
+  # }}}
+  # {{{ os health
+  ####
+  -
+    name: users
+    nrpe: "/usr/lib/nagios/plugins/check_users 30 35"
+    hostgroups: computers
+  -
+    name: load
+    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
+  -
+    name: uptime check
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-uptime"
+    hostgroups: computers
+  -
+    name: processes - total
+    nrpe: "/usr/lib/nagios/plugins/check_procs 620 700"
+    hostgroups: computers
+  -
+    name: swap usage - percent
+    nrpe: "/usr/lib/nagios/plugins/check_swap -w 20% -c 10%"
+    hostgroups: computers
+  -
+    name: swap usage - mb
+    nrpe: "/usr/lib/nagios/plugins/check_swap -w 20000 -c 5000"
+    hostgroups: computers
+  -
+    name: process - getty
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C getty -a /sbin/getty"
+    hostgroups: computers
+    excludehosts: zelenka, zandonai
+  -
+    name: processes - zombies
+    nrpe: "/usr/lib/nagios/plugins/check_procs 5 10 -s Z"
+    hostgroups: computers
+  -
+    name: system - available entropy
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-entropy"
+    event_handler: dsa_event_handler_restart_ekey
     hostgroups: computers
+    excludehostgroups: freebsd
+  -
+    name: system - filesystem check
+    nrpe: "/usr/bin/sudo /usr/lib/nagios/plugins/dsa-check-filesystems"
     normal_check_interval:  60
     retry_check_interval: 15
-  ####
+    hostgroups: computers
+  # }}}
+  # {{{ backup
   -
     name: backup
     servicegroups: backup
@@ -1471,6 +1521,12 @@ services:
     hostgroups: freebsd
 
   ####
+  -
+    name: process - acc.umu.se backup
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -a 'dsmc'"
+    hosts: sibelius
+  # }}}
+  # {{{ security
   -
     name: running kernel
     servicegroups: kernel
@@ -1478,8 +1534,13 @@ services:
     hostgroups: computers
     normal_check_interval: 60
     retry_check_interval: 5
-
-  ####
+  -
+    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: installed firewall
     nrpe: "/usr/lib/nagios/plugins/dsa-check-file -w -f /etc/ferm/ferm.conf"
@@ -1490,12 +1551,15 @@ services:
     nrpe: "/usr/lib/nagios/plugins/dsa-check-file -w -f /etc/ferm/conf.d/defs.conf"
     hostgroups: computers
     excludehostgroups: freebsd
-
-  ####
-  - name: ganeti - job watcher paused
-    nrpe: "/usr/lib/nagios/plugins/negate /usr/lib/nagios/plugins/dsa-check-file -f /var/lib/ganeti/watcher.pause"
+  -
+    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: computers
-
+    excludehostgroups: freebsd, sparc
+  -
+    name: unexpected process - ulogd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C ulogd"
+    hostgroups: freebsd, sparc
   ####
   -
     name: process - samhain
@@ -1511,74 +1575,39 @@ services:
     normal_check_interval: 60
     retry_check_interval: 5
     excludehostgroups: brokensamhain
-
-  ####
-  -
-    name: process - acc.umu.se backup
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -a 'dsmc'"
-    hosts: sibelius
-
-  ####
-  -
-    name: users
-    nrpe: "/usr/lib/nagios/plugins/check_users 30 35"
-    hostgroups: computers
-
-  ####
-  -
-    name: load
-    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
-
-  ####
-  -
-    name: uptime check
-    nrpe: "/usr/lib/nagios/plugins/dsa-check-uptime"
-    hostgroups: computers
-
-  ####
   -
     name: processes - samhain zombies
     nrpe: "/usr/lib/nagios/plugins/check_procs 3 6 -s Z -u root -a samhain"
     event_handler: dsa_event_handler_restart_samhain
     hostgroups: computers
     excludehostgroups: brokensamhain
+  # }}}
+  # {{{ logging
   -
-    name: processes - zombies
-    nrpe: "/usr/lib/nagios/plugins/check_procs 5 10 -s Z"
-    hostgroups: computers
-
-  ####
-  -
-    name: processes - total
-    nrpe: "/usr/lib/nagios/plugins/check_procs 620 700"
+    name: process - syslog-ng
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C syslog-ng  -a '/sbin/syslog-ng -p /var/run/syslog-ng.pid'"
     hostgroups: computers
+    excludehostgroups: freebsd
 
-  ####
   -
-    name: swap usage - percent
-    nrpe: "/usr/lib/nagios/plugins/check_swap -w 20% -c 10%"
-    hostgroups: computers
+    name: process - syslog-ng
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 2: -u root -C syslog-ng  -a '/sbin/syslog-ng -p /var/run/syslog-ng.pid'"
+    hostgroups: freebsd
 
-  ####
   -
-    name: swap usage - mb
-    nrpe: "/usr/lib/nagios/plugins/check_swap -w 20000 -c 5000"
+    name: remote logging on lotti
+    remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
+    runfrom: lotti
     hostgroups: computers
-
-  ####
+    excludehostgroups: alioth
   -
-    name: process - getty
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C getty -a /sbin/getty"
+    name: remote logging on lully
+    remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
+    runfrom: lully
     hostgroups: computers
-    excludehosts: zelenka, zandonai
-
-  ####
+    excludehostgroups: alioth
+  # }}}
+  # {{{ base service
   -
     name: process - sshd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:20 -c 1: -u root -C sshd -a '/usr/sbin/sshd'"
@@ -1590,7 +1619,6 @@ services:
     depends: process - sshd
     normal_check_interval:  60
     notification_interval: 1440
-
   ####
   -
     name: network service - nrpe
@@ -1604,7 +1632,6 @@ services:
     hostgroups: computers
     max_check_attempts: -1
     depends: network service - nrpe
-
   ###
   -
     name: process - munin-node
@@ -1625,7 +1652,6 @@ services:
     check: check_tcp!4949
     hostgroups: computers
     depends: process - munin-node
-
   ###
   -
     name: process - ntpd
@@ -1639,7 +1665,6 @@ services:
     depends: process - ntpd
     excludehosts: ancina
     servicegroups: time
-
   ###
   -
     name: network service - time
@@ -1647,7 +1672,6 @@ services:
     hosts: ancina
     depends: process - xinetd
     servicegroups: time
-
   ###
   -
     name: process - atd
@@ -1658,31 +1682,27 @@ services:
     name: process - cron
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u root -C cron -a /usr/sbin/cron"
     hostgroups: computers
-
   ###
   -
-    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'"
+    name: process - ud-replicated
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C ud-replicated -a '/usr/bin/python /usr/bin/ud-replicated'"
     hostgroups: computers
-    excludehostgroups: freebsd
-
+    excludehostgroups: squeeze, freebsd, alioth
   -
-    name: process - syslog-ng
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:2 -c 2: -u root -C syslog-ng  -a '/sbin/syslog-ng -p /var/run/syslog-ng.pid'"
+    name: process - ud-replicated
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C python2.7 -a '/usr/bin/python /usr/bin/ud-replicated'"
     hostgroups: freebsd
-
+  ###
   -
-    name: remote logging on lotti
-    remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
-    runfrom: lotti
-    hostgroups: computers
-    excludehostgroups: alioth
+    name: process - monit
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/sbin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
+    hostgroups: squeeze
   -
-    name: remote logging on lully
-    remotecheck: "/usr/lib/nagios/plugins/dsa-check-log-age-loghost $HOSTNAME$"
-    runfrom: lully
+    name: process - monit
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/bin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
     hostgroups: computers
-    excludehostgroups: alioth
+    excludehostgroups: squeeze, alioth
+  ###
   -
     name: MQ connection on rainier
     remotecheck: "/usr/lib/nagios/plugins/dsa-check-mq-connection $HOSTNAME$ ud dsa"
@@ -1699,276 +1719,73 @@ services:
     normal_check_interval:  60
     retry_check_interval: 15
     excludehostgroups: alioth
-  # }}}
-  # {{{ ############ MAIL STUFF ############
+  ###
   -
-    name: process - exim
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
+    name: local resolver
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-resolver www.debian.org www.google.com"
     hostgroups: computers
-    excludehostgroups: postfix-hosts, mail-relay
-    excludehosts: master, busoni, quantz, buxtehude
+    normal_check_interval: 60
   -
-    name: process - exim
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
-    hostgroups: mail-relay
+    name: process - unbound
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u unbound -C unbound -a '/usr/sbin/unbound'"
+    hostgroups: computers
+    excludehostgroups: alioth
+  ###
   -
-    name: process - exim - total
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1: -C exim4"
+    name: process - uptimed
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u daemon -C uptimed -a '/usr/sbin/uptimed'"
     hostgroups: computers
-    excludehostgroups: postfix-hosts
-    excludehosts: master, busoni, quantz, buxtehude
+  ###
   -
-    name: process - exim
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:300 -c 1:500 -C exim4 -a '/usr/sbin/exim4'"
-    hosts: master, busoni, quantz, buxtehude
+    name: process - udevd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -p 1 -C udevd -a 'udevd'"
+    hostgroups: computers
+    excludehostgroups: freebsd
   -
-    name: mail queue
-    nrpe: "/usr/lib/nagios/plugins/check_mailq -M exim -w 1000 -c 2000"
-    hostgroups: heavy-exim
+    name: unexpected process - udev
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C udevd"
+    hostgroups: freebsd
   ###
   -
-    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-exim, heavy-postfix
+    name: process - acpid
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C acpid -a '/usr/sbin/acpid'"
+    hostgroups: acpid-hosts
   -
-    name: service - clamav
-    nrpe: "/usr/lib/nagios/plugins/check_clamd -H /var/run/clamav/clamd.ctl"
-    hostgroups: heavy-exim, heavy-postfix
-    depends: process - clamav - clamd
+    name: unexpected process - acpid
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C acpid"
+    hostgroups: computers
+    excludehostgroups: acpid-hosts, kvmdomains
+  ###
   -
-    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-exim, heavy-postfix
-  -
-    name: unwanted process - clamav
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C clamd"
-    hostgroups: computers
-    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-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'"
-    hostgroups: spamd
-    excludehosts: picconi
-  -
-    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'"
-    hosts: picconi
-  -
-    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: bendel
-  -
-    name: process - spamd - child
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:11 -c 1: -C spamd -a 'spamd child'"
-    hosts: bendel
-    hostgroups: spamd
-    depends: process - spamd - master
-  #
-  -
-    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: 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: buxtehude
-  #
-  -
-    name: unwanted process - spamd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd"
-    hostgroups: computers
-    excludehostgroups: spamd
-    excludehosts: bendel, busoni, buxtehude
-
-  -
-    name: unwanted process - greylistd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C greylistd"
-    hostgroups: computers
-
-  ###
-  -
-    name: process - postgrey
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --unix=/var/run/postgrey/socket --retry-window=4 --auto-whitelist-clients=10 --exim'"
-    hostgroups: heavy-exim
-  -
-    name: process - postgrey
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --inet=127.0.0.1:60000'"
-    hostgroups: heavy-postfix
-  #
+    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'"
+    hostgroups: xinetd-hosts
   -
-    name: unwanted process - postgrey
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C postgrey"
+    name: unwanted process - xinetd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C xinetd"
     hostgroups: computers
-    excludehostgroups: heavy-postfix, heavy-exim
+    excludehostgroups: xinetd-hosts
   ###
   -
-    name: process - amavis - master
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u amavis -a 'amavisd-new (master)'"
-    hostgroups: amavis-hosts
-  -
-    name: process - amavis - all
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1:10 -u amavis -a 'amavisd-new '"
-    hostgroups: amavis-hosts
-    depends: process - amavis - master
-  #
+    name: process - stunnel4 - puppet-ekeyd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1: -u stunnel4 -C stunnel4 -a '/usr/bin/stunnel4 /etc/stunnel/puppet-ekeyd.conf'"
+    hostgroups: squeeze, wheezy
+    excludehostgroups: freebsd, alioth
+  # }}}
+  # {{{ anti-services
   -
-    name: unwanted process - amavis
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C amavisd-new"
+    name: ganeti - job watcher paused
+    nrpe: "/usr/lib/nagios/plugins/negate /usr/lib/nagios/plugins/dsa-check-file -f /var/lib/ganeti/watcher.pause"
     hostgroups: computers
-    excludehostgroups: amavis-hosts
-  ###
   -
-    name: process - weightd - master
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (master)'"
-    hostgroups: heavy-postfix, alioth
-  -
-    name: process - weightd - cache
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (cache)'"
-    hostgroups: heavy-postfix, alioth
-    depends: process - weightd - master
-  -
-    name: process - weightd - child
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:50 -c 1: -u polw -a 'policyd-weight (child)'"
-    hostgroups: heavy-postfix, alioth
-    depends: process - weightd - master
-  ###
-  -
-    name: unwanted process - policyd-weight
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C policyd-weight"
+    name: unwanted process - irqbalance
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance"
     hostgroups: computers
-    excludehostgroups: heavy-postfix, alioth
-
-
-  ###
-  -
-    name: process - postfix - master
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C master -a '/usr/lib/postfix/master'"
-    hostgroups: postfix-hosts
-  -
-    name: process - postfix - qmgr
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C qmgr -a 'qmgr -l -t fifo -u'"
-    hostgroups: postfix-hosts
-    depends: process - postfix - master
-  -
-    name: process - postfix - pickup
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C pickup -a 'pickup -l -t fifo -u -c'"
-    hostgroups: postfix-hosts
-    depends: process - postfix - master
-  -
-    name: process - postfix - anvil
-    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
-
-  -
-    name: process - postfix - trivial-rewrite
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C trivial-rewrite -a 'trivial-rewrite -n rewrite -t unix -u -c'"
-    hostgroups: postfix-hosts
-    depends: process - postfix - master
-  -
-    name: process - postfix - proxymap
-    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 - 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'"
-    hostgroups: postfix-hosts
-    depends: process - postfix - master
-  -
-    name: process - postfix - local
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u postfix -C local -a 'local -t unix'"
-    hostgroups: postfix-hosts
-    depends: process - postfix - master
   -
     name: unwanted process - openvpn
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C openvpn"
     hostgroups: computers
     normal_check_interval: 120
-
-
-  ###
-  -
-    name: network service - smtp
-    check: dsa_check_smtp
-    hostgroups: computers
-    excludehostgroups: postfix-hosts, incomingmailrelayed, incomingmailrelayed2025
-    depends: process - exim
-
-  -
-    name: network service - smtp
-    check: dsa_check_smtp
-    hostgroups: postfix-hosts
-    depends: process - postfix - master
-  -
-    name: network service - submission
-    check: dsa_check_smtp_port!587
-    hostgroups: incomingmailrelayed
-    depends: process - exim
-  -
-    name: network service - smtp 2025
-    check: dsa_check_smtp_port!2025
-    hostgroups: incomingmailrelayed2025
-    depends: process - exim
-  -
-    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: 60
-    excludehostgroups: alioth
-  -
-    name: setup - local hostname etc-hosts
-    nrpe: 'if getent ahosts `hostname` | grep -q 127.0; then echo "Warning: local hostname resolves to 127/8 address"; exit 1; else echo "OK: Hostname resolves to non-127/8 address."; exit 0; fi'
-    hostgroups: computers
-    normal_check_interval: 60
-  -
-    name: system - available entropy
-    nrpe: "/usr/lib/nagios/plugins/dsa-check-entropy"
-    event_handler: dsa_event_handler_restart_ekey
-    hostgroups: computers
-    excludehostgroups: freebsd
-  -
-    name: system - filesystem check
-    nrpe: "/usr/bin/sudo /usr/lib/nagios/plugins/dsa-check-filesystems"
-    normal_check_interval:  60
-    retry_check_interval: 15
-    hostgroups: computers
-  ###
-  -
-    name: local resolver
-    nrpe: "/usr/lib/nagios/plugins/dsa-check-resolver www.debian.org www.google.com"
-    hostgroups: computers
-    normal_check_interval: 60
-  -
-    name: process - unbound
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u unbound -C unbound -a '/usr/sbin/unbound'"
-    hostgroups: computers
-    excludehostgroups: alioth
-  ###
-  -
-    name: process - uptimed
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u daemon -C uptimed -a '/usr/sbin/uptimed'"
-    hostgroups: computers
-  ###
-  -
-    name: unwanted process - irqbalance
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance"
-    hostgroups: computers
-
-  ####
   -
     name: unwanted process - gkrellmd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C gkrellmd"
@@ -1987,19 +1804,14 @@ services:
     name: unwanted process - snmpd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C snmpd"
     hostgroups: computers
-
-  ####
+  # }}}
+  # {{{ ssl certs
   -
     name: "host SSL cert"
     nrpe: "if [ -e /etc/ssl/certs/thishost.pem ]; then /usr/lib/nagios/plugins/dsa-check-cert-expire /etc/ssl/certs/thishost.pem; else echo 'No thishost.pem on this host.'; fi"
     hostgroups: computers
   # }}}
-  # {{{ ############ Processes/Services that only run on some computers ############
-  -
-    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'"
-    hostgroups: dl385
-  ###
+  # {{{ HW health/raid
   -
     name: process - mdadm monitor
     servicegroups: raid
@@ -2018,26 +1830,7 @@ services:
     servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-sw"
     hostgroups: sw-raid
-
   ###
-  -
-    name: process - ud-replicated
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C ud-replicated -a '/usr/bin/python /usr/bin/ud-replicated'"
-    hostgroups: computers
-    excludehostgroups: squeeze, freebsd, alioth
-  -
-    name: process - ud-replicated
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C python2.7 -a '/usr/bin/python /usr/bin/ud-replicated'"
-    hostgroups: freebsd
-  -
-    name: process - monit
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/sbin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
-    hostgroups: squeeze
-  -
-    name: process - monit
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/bin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
-    hostgroups: computers
-    excludehostgroups: squeeze, alioth
   -
     name: HW - hpacucli status
     servicegroups: raid
@@ -2117,107 +1910,238 @@ services:
     servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/dsa-check-raid-mpt"
     hostgroups: mptraid
-
   ###
   -
     name: RAID - DRBD
     servicegroups: raid
     nrpe: "/usr/lib/nagios/plugins/dsa-check-drbd -d All"
     hostgroups: drbd-hosts
-  ###
+  # }}}
+  # }}}
+  # {{{ ### mail stuff
+  # {{{ exim processes and mailq
   -
-    name: process - slapd
-    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: draghi
+    name: process - exim
+    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, mail-relay
+    excludehosts: master, busoni, quantz, buxtehude
   -
-    name: network service - ldaps cert
-    check: dsa_check_cert!636
-    depends: process - slapd
-    normal_check_interval: 60
-    hosts: draghi
-  ###
+    name: process - exim
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:25 -c 1: -u Debian-exim -C exim4 -a '/usr/sbin/exim4 -bd -q'"
+    hostgroups: mail-relay
   -
-    name: process - ulogd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C ulogd -a '/usr/sbin/ulogd -d'"
+    name: process - exim - total
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1: -C exim4"
     hostgroups: computers
-    excludehostgroups: freebsd, sparc
+    excludehostgroups: postfix-hosts
+    excludehosts: master, busoni, quantz, buxtehude
   -
-    name: unexpected process - ulogd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C ulogd"
-    hostgroups: freebsd, sparc
+    name: process - exim
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:300 -c 1:500 -C exim4 -a '/usr/sbin/exim4'"
+    hosts: master, busoni, quantz, buxtehude
+  -
+    name: mail queue
+    nrpe: "/usr/lib/nagios/plugins/check_mailq -M exim -w 1000 -c 2000"
+    hostgroups: heavy-exim
+  # }}}
+  # {{{ clamav
+  -
+    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-exim, heavy-postfix
+  -
+    name: service - clamav
+    nrpe: "/usr/lib/nagios/plugins/check_clamd -H /var/run/clamav/clamd.ctl"
+    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-exim, heavy-postfix
+  -
+    name: unwanted process - clamav
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C clamd"
+    hostgroups: computers
+    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-exim, heavy-postfix
+  # }}}
+  # {{{ anti-spam
+  -
+    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: picconi
+  -
+    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'"
+    hosts: picconi
+  -
+    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: bendel
+  -
+    name: process - spamd - child
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:11 -c 1: -C spamd -a 'spamd child'"
+    hosts: bendel
+    hostgroups: spamd
+    depends: process - spamd - master
+  #
+  -
+    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: 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: buxtehude
+  #
+  -
+    name: unwanted process - spamd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd"
+    hostgroups: computers
+    excludehostgroups: spamd
+    excludehosts: bendel, busoni, buxtehude
+
+  -
+    name: unwanted process - greylistd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C greylistd"
+    hostgroups: computers
 
   ###
   -
-    name: process - udevd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -p 1 -C udevd -a 'udevd'"
+    name: process - postgrey
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a '/usr/sbin/postgrey --pidfile=/var/run/postgrey.pid --daemonize --unix=/var/run/postgrey/socket --retry-window=4 --auto-whitelist-clients=10 --exim'"
+    hostgroups: heavy-exim
+  -
+    name: process - postgrey
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u 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: freebsd
+    excludehostgroups: heavy-postfix, heavy-exim
+  ###
   -
-    name: unexpected process - udev
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C udevd"
-    hostgroups: freebsd
+    name: process - amavis - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u amavis -a 'amavisd-new (master)'"
+    hostgroups: amavis-hosts
+  -
+    name: process - amavis - all
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1:10 -u amavis -a 'amavisd-new '"
+    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 - acpid
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C acpid -a '/usr/sbin/acpid'"
-    hostgroups: acpid-hosts
+    name: process - weightd - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (master)'"
+    hostgroups: heavy-postfix, alioth
+  -
+    name: process - weightd - cache
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u polw -a 'policyd-weight (cache)'"
+    hostgroups: heavy-postfix, alioth
+    depends: process - weightd - master
   -
-    name: unexpected process - acpid
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C acpid"
-    hostgroups: computers
-    excludehostgroups: acpid-hosts, kvmdomains
+    name: process - weightd - child
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:50 -c 1: -u polw -a 'policyd-weight (child)'"
+    hostgroups: heavy-postfix, alioth
+    depends: process - weightd - master
   ###
   -
-    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'"
-    hostgroups: xinetd-hosts
-  -
-    name: unwanted process - xinetd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C xinetd"
+    name: unwanted process - policyd-weight
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C policyd-weight"
     hostgroups: computers
-    excludehostgroups: xinetd-hosts
+    excludehostgroups: heavy-postfix, alioth
+  # }}}
+  # {{{ postfix
   ###
   -
-    name: network service - finger
-    check: check_tcp!79
-    hosts: draghi
-    depends: process - xinetd
-  ###
+    name: process - postfix - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C master -a '/usr/lib/postfix/master'"
+    hostgroups: postfix-hosts
   -
-    name: network service - rsync
-    check: check_tcp!873
-    hostgroups: rsyncd-hosts
-    depends: process - xinetd
-    excludehosts: rietz
+    name: process - postfix - qmgr
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C qmgr -a 'qmgr -l -t fifo -u'"
+    hostgroups: postfix-hosts
+    depends: process - postfix - master
   -
-    name: network service - rsync
-    check: check_tcp!873
-    hosts: rietz2
-    depends: rietz:process - xinetd
+    name: process - postfix - pickup
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postfix -C pickup -a 'pickup -l -t fifo -u -c'"
+    hostgroups: postfix-hosts
+    depends: process - postfix - master
   -
-    name: network service - rsync
-    check: check_tcp!873
-    hosts: milanollo2
-    depends: milanollo:process - xinetd
+    name: process - postfix - anvil
+    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
 
-  ###
   -
-    name: process - icinga
-    # 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 icinga -a '/usr/sbin/icinga -d /etc/icinga/icinga.cfg'"
-    hosts: tchaikovsky
-  ###
+    name: process - postfix - trivial-rewrite
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C trivial-rewrite -a 'trivial-rewrite -n rewrite -t unix -u -c'"
+    hostgroups: postfix-hosts
+    depends: process - postfix - master
   -
-    name: process - jetty - master
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -a 'jsvc.exec'"
-    hostgroups: jetty-hosts
+    name: process - postfix - proxymap
+    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 - jetty - worker
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1:100 -u jetty -a 'jsvc.exec -user jetty'"
-    hostgroups: jetty-hosts
-    depends: process - jetty - 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'"
+    hostgroups: postfix-hosts
+    depends: process - postfix - master
+  -
+    name: process - postfix - local
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u postfix -C local -a 'local -t unix'"
+    hostgroups: postfix-hosts
+    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
+  # }}}
+  # {{{ mail - network service
+  -
+    name: network service - smtp
+    check: dsa_check_smtp
+    hostgroups: computers
+    excludehostgroups: postfix-hosts, incomingmailrelayed, incomingmailrelayed2025
+    depends: process - exim
 
-  ###
+  -
+    name: network service - smtp
+    check: dsa_check_smtp
+    hostgroups: postfix-hosts
+    depends: process - postfix - master
+  -
+    name: network service - submission
+    check: dsa_check_smtp_port!587
+    hostgroups: incomingmailrelayed
+    depends: process - exim
+  -
+    name: network service - smtp 2025
+    check: dsa_check_smtp_port!2025
+    hostgroups: incomingmailrelayed2025
+    depends: process - exim
+  # }}}
+  # }}}
+  # {{{ ### host specific services
+  # {{{ HTTP
   -
     name: process - apache2 - master
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -a /usr/sbin/apache2"
@@ -2287,24 +2211,8 @@ services:
     # ravel does ssh on port 443
     excludehosts: ravel
     normal_check_interval: 60
-
-  ####
-  -
-    name: process - named
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:15 -c 1: -u bind -C named -a '/usr/sbin/named -u bind'"
-    hostgroups: bind9-hosts
-  -
-    name: network service - dns
-    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
-
-  ####
+  # }}}
+  # {{{ FTP
   -
     name: network service - ftp
     check: check_ftp
@@ -2315,14 +2223,8 @@ services:
     check: check_ftp
     hosts: klecker-ftp
     depends: klecker:process - xinetd
-
-  -
-    name: process - debianqueued
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u dak-unpriv -C debianqueued"
-    hostgroups: uploadqueue
-
-
-  ####
+  # }}}
+  # {{{ postgres
   -
     name: unwanted process - postgresql
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C postgres"
@@ -2340,13 +2242,8 @@ services:
     name: postgresql backups
     nrpe: "/usr/bin/sudo -u debbackup /usr/lib/nagios/plugins/dsa-check-backuppg"
     hosts: backuphost
-  ####
-  -
-    name: process - stunnel4 - puppet-ekeyd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1: -u stunnel4 -C stunnel4 -a '/usr/bin/stunnel4 /etc/stunnel/puppet-ekeyd.conf'"
-    hostgroups: squeeze, wheezy
-    excludehostgroups: freebsd, alioth
-  ####
+  # }}}
+  # {{{ 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'"
@@ -2369,7 +2266,8 @@ services:
     nrpe: "/usr/lib/nagios/plugins/dsa-check-ups"
     hosts: franck
     depends: process - UPS - nut upsd
-  ###
+  # }}}
+  # {{{ buildd
   -
     name: process - buildd
     servicegroups: buildd
@@ -2396,14 +2294,8 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_procs -m 'ELAPSED' -c 500 -C lvcreate -u root -a 'lvcreate'"
     hostgroups: buildd
     contact_groups: +buildd
-  ####
-  -
-    name: network service - gobby
-    check: check_tcp!6523
-    hosts: gombert
-    contact_groups: gobby
   # }}}
-  # {{{ ############ NFS Stuff ############
+  # {{{ NFS Stuff
   -
     name: process - statd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u statd -C rpc.statd -a '/sbin/rpc.statd'"
@@ -2431,14 +2323,7 @@ services:
     hostgroups: computers
     excludehostgroups: autofs
   # }}}
-  # {{{ ############ MISC OTHER Stuff ############
-  -
-    name: puppetmaster cert
-    nrpe: "/usr/lib/nagios/plugins/dsa-check-cert-expire /var/lib/puppet/ssl/certs/ca.pem"
-    hosts: handel
-    normal_check_interval: 60
-    max_check_attempts: 2
-    retry_check_interval: 5
+  # {{{ mirroring
   -
     name: mirror sync - bugs
     check: "dsa_check_mirrorsync_skew!bugs.debian.org!project/trace/bugs-master.debian.org!120:600"
@@ -2461,6 +2346,23 @@ services:
     normal_check_interval: 15
     max_check_attempts: 5
     retry_check_interval: 5
+  # }}}
+  # {{{ DNS
+  -
+    name: process - named
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:15 -c 1: -u bind -C named -a '/usr/sbin/named -u bind'"
+    hostgroups: bind9-hosts
+  -
+    name: network service - dns
+    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
+  ###
   -
     name: DNS SOA sync - debian.org
     check: "dsa_check_soas_add!denis.debian.org!debian.org"
@@ -2501,8 +2403,8 @@ services:
     hosts: global
     remotecheck: "/usr/lib/nagios/plugins/dsa-check-statusfile /srv/dns.debian.org/var/nagios/ds"
     runfrom: denis
-
-  ############
+  # }}}
+  # {{{ storage
   -
     name: ping alive check
     remotecheck: "/usr/lib/nagios/plugins/check_ping -H $HOSTADDRESS$ -w 50,10% -c 200,30%"
@@ -2521,7 +2423,8 @@ services:
     remotecheck: "/usr/lib/nagios/plugins/dsa-check-msa-eventlog --start=8867 $HOSTADDRESS$ public"
     runfrom: dijkstra
     hosts: giustini
-  ############
+  # }}}
+  # {{{ porterbox
   -
     name: current chroots
     nrpe: "/usr/lib/nagios/plugins/dsa-check-dchroots-current"
@@ -2529,7 +2432,7 @@ services:
     normal_check_interval:  60
     retry_check_interval: 15
   # }}}
-  # {{{ ############ openstack ############
+  # {{{ openstack
   -
     name: process - openstack - memcached
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u nobody -C memcached -a '/usr/bin/memcached -m 64 -p 11211 -u nobody -l 127.0.0.1'"
@@ -2571,6 +2474,80 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u neutron -C neutron-server -a '/usr/bin/python2.7 /usr/bin/neutron-server --config-file=/etc/neutron/neutron.conf --config-file /etc/neutron/plugins/openvswitch/ovs_neutron_plugin.ini --log-file=/var/log/neutron/neutron-server.log'"
     hostgroups: openstack-controller
   # }}}
+  # {{{ misc
+  -
+    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'"
+    hostgroups: dl385
+  ###
+  -
+    name: process - slapd
+    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: draghi
+  -
+    name: network service - ldaps cert
+    check: dsa_check_cert!636
+    depends: process - slapd
+    normal_check_interval: 60
+    hosts: draghi
+  ###
+  -
+    name: network service - finger
+    check: check_tcp!79
+    hosts: draghi
+    depends: process - xinetd
+  ###
+  -
+    name: network service - rsync
+    check: check_tcp!873
+    hostgroups: rsyncd-hosts
+    depends: process - xinetd
+    excludehosts: rietz
+  -
+    name: network service - rsync
+    check: check_tcp!873
+    hosts: rietz2
+    depends: rietz:process - xinetd
+  -
+    name: network service - rsync
+    check: check_tcp!873
+    hosts: milanollo2
+    depends: milanollo:process - xinetd
+  ###
+  -
+    name: process - icinga
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:30 -c 1: -u nagios -C icinga -a '/usr/sbin/icinga -d /etc/icinga/icinga.cfg'"
+    hosts: tchaikovsky
+  ###
+  -
+    name: process - jetty - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -a 'jsvc.exec'"
+    hostgroups: jetty-hosts
+  -
+    name: process - jetty - worker
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1:100 -u jetty -a 'jsvc.exec -user jetty'"
+    hostgroups: jetty-hosts
+    depends: process - jetty - master
+  ###
+  -
+    name: process - debianqueued
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:3 -c 1: -u dak-unpriv -C debianqueued"
+    hostgroups: uploadqueue
+  ####
+  -
+    name: network service - gobby
+    check: check_tcp!6523
+    hosts: gombert
+    contact_groups: gobby
+  ####
+  -
+    name: puppetmaster cert
+    nrpe: "/usr/lib/nagios/plugins/dsa-check-cert-expire /var/lib/puppet/ssl/certs/ca.pem"
+    hosts: handel
+    normal_check_interval: 60
+    max_check_attempts: 2
+    retry_check_interval: 5
+  # }}}
 # }}}
 
 # vim: set ts=2 sw=2 et ai si fdm=marker:
-- 
2.20.1