X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=config%2Fnagios-master.cfg;h=e9c134c388d58a68387cca25a9f7c56ef217981d;hb=cc33b9c07688eabb479083349004b2720599f379;hp=ba445cccf7eb848a56852f0cafc7904cd464b6b7;hpb=c4643a69f0b67bf2b2d503236b72d36a178e45e2;p=mirror%2Fdsa-nagios.git

diff --git a/config/nagios-master.cfg b/config/nagios-master.cfg
index ba445cc..e9c134c 100644
--- a/config/nagios-master.cfg
+++ b/config/nagios-master.cfg
@@ -139,7 +139,7 @@ servers:
     parents: gw-ubcece
     hostgroups: layer3-infrastructure
   gw-unicamp:
-    address: 177.220.10.129
+    address: 143.106.167.113
     parents: gw-ubcece
     hostgroups: layer3-infrastructure
   gw-utwente:
@@ -238,15 +238,15 @@ servers:
   arm-arm-01:
     address: 217.140.96.58
     parents: gw-arm
-    hostgroups: computers, hassrvfs, buildd, stretch, broken_mq
+    hostgroups: computers, hassrvfs, buildd, stretch, broken_mq, sw-raid
   arm-arm-03:
     address: 217.140.96.60
     parents: gw-arm
-    hostgroups: computers, hassrvfs, buildd, stretch, broken_mq
+    hostgroups: computers, hassrvfs, buildd, stretch, broken_mq, sw-raid
   arm-arm-04:
     address: 217.140.96.61
     parents: gw-arm
-    hostgroups: computers, hassrvfs, buildd, stretch, broken_mq
+    hostgroups: computers, hassrvfs, buildd, stretch, broken_mq, sw-raid
   harris:
     address: 217.140.96.66
     parents: gw-arm
@@ -405,9 +405,7 @@ servers:
   moszumanska:
     address: 5.153.231.21
     parents: ganeti-bytemark
-    contact_groups: alioth-admins
-    hostgroups: computers, general, wheezy, postgres91-hosts, apache2-hosts, acpid-hosts, apache-https, brokensamhain, no-bacula, bind9-hosts, xinetd-hosts, alioth, heavy-exim, spamd
-    no-servicegroups: true
+    hostgroups: secondary-IPs
   dillon:
     address: 5.153.231.22
     parents: ganeti-bytemark
@@ -593,7 +591,7 @@ servers:
   pkgmirror-csail:
     address: 128.31.0.51
     parents: ganeti-csail
-    hostgroups: computers, service, kvmdomains, stretch, apache2-hosts, no-bacula, apache-https, hassrvfs, systemd-timesyncd
+    hostgroups: computers, service, kvmdomains, stretch, apache2-hosts, no-bacula, apache-https, hassrvfs, systemd-timesyncd, varnish-hosts
   usper:
     address: 128.31.0.69
     parents: ganeti-csail
@@ -683,27 +681,32 @@ servers:
   lw01:
     address: 185.17.185.177
     parents: gw-leaseweb
-    hostgroups: computers, service, jessie, dl180, nfs-server, rsyncd-hosts
+    hostgroups: computers, service, stretch, dl180, nfs-server, rsyncd-hosts
   lw02:
     address: 185.17.185.178
     parents: gw-leaseweb
-    hostgroups: computers, service, jessie, dl180, nfs-server, rsyncd-hosts
+    hostgroups: computers, service, stretch, dl180, nfs-server, rsyncd-hosts
   lw03:
     address: 185.17.185.179
     parents: gw-leaseweb
-    hostgroups: computers, service, jessie, dl180, nfs-server, rsyncd-hosts
+    hostgroups: computers, service, stretch, dl180, nfs-server, rsyncd-hosts
   lw04:
     address: 185.17.185.180
     parents: gw-leaseweb
-    hostgroups: computers, service, jessie, dl180, nfs-server, rsyncd-hosts
+    hostgroups: computers, service, stretch, dl180, nfs-server, rsyncd-hosts
   lw07:
     address: 185.17.185.187
     parents: gw-leaseweb
-    hostgroups: computers, service, jessie, dl180, nfs-client, autofs, hassrvfs, postgres94-hosts, apache2-hosts
+    hostgroups: computers, service, stretch, dl180, nfs-client, autofs, hassrvfs, postgres96-hosts, apache2-hosts, haproxy-hosts, haproxy-https-host, varnish-hosts
+  lw07-2:
+    address: 185.17.185.185
+    parents: lw07
+    hostgroups: secondary-IPs, https-service
+
   lw08:
     address: 185.17.185.189
     parents: gw-leaseweb
-    hostgroups: computers, service, jessie, dl180, nfs-client, autofs, hassrvfs, apache2-hosts
+    hostgroups: computers, service, stretch, dl180, nfs-client, autofs, hassrvfs, apache2-hosts
   lw09:
     address: 185.17.185.181
     parents: gw-leaseweb
@@ -831,25 +834,20 @@ servers:
     address: 140.211.166.197
     parents: pieta
     hostgroups: computers, hassrvfs, buildd, jessie
-  partch:
-    address: 140.211.15.152
-    parents: gw-osuosl
-    hostgroups: computers, jessie, hassrvfs, porterbox, sw-raid
   # }}}
   # {{{ gw-sanger
   sallinen:
     address: 193.62.202.26
     parents: gw-sanger
-    hostgroups: computers, service, stretch, dl380, nfs-client, autofs, postgres96-hosts
+    hostgroups: computers, service, stretch, dl380, nfs-client, autofs, postgres96-hosts, apache2-hosts, haproxy-hosts, haproxy-https-host, varnish-hosts
+  sallinen-2:
+    address: 193.62.202.27
+    parents: sallinen
+    hostgroups: secondary-IPs, https-service
   sibelius:
     address: 193.62.202.28
     parents: gw-sanger
-    hostgroups: computers, postgres94-hosts, service, apache2-hosts, sw-raid, jessie, rsyncd-hosts, hasvarlogfs, multipath-hosts, nfs-server
-    contacts: tjrc1, dave
-  smetana:
-    address: 193.62.202.29
-    parents: gw-sanger
-    hostgroups: computers, sw-raid, sparc, wheezy, no-bacula
+    hostgroups: computers, postgres94-hosts, service, apache2-hosts, sw-raid, jessie, rsyncd-hosts, hasvarlogfs, multipath-hosts, nfs-server, varnish-hosts
     contacts: tjrc1, dave
   # }}}
   # {{{ gw-scanplus
@@ -989,6 +987,14 @@ servers:
     address: 209.87.16.46
     parents: ubc-gateway
     hostgroups: computers, service, kvmdomains, stretch, systemd-timesyncd, apache2-hosts, apache-https, broken_https_default_vhost
+  kantuser:
+    address: 209.87.16.47
+    parents: ubc-gateway
+    hostgroups: computers, service, kvmdomains, stretch, systemd-timesyncd, apache2-hosts
+  grabbe:
+    address: 209.87.16.48
+    parents: ubc-gateway
+    hostgroups: computers, service, kvmdomains, stretch, systemd-timesyncd, apache2-hosts, apache-https
   # }}}
   # {{{ gw-umn
   #saens:
@@ -1014,19 +1020,19 @@ servers:
   # }}}
   # {{{ gw-unicamp
   prokofiev:
-    address: 177.220.10.140
+    address: 143.106.167.124
     parents: gw-unicamp
     hostgroups: computers, stretch, service, manyprocesses
   powerpc-unicamp-01:
-    address: 177.220.10.141
+    address: 143.106.167.120
     parents: prokofiev
     hostgroups: computers, hassrvfs, buildd, jessie
   ppc64el-unicamp-01:
-    address: 177.220.10.142
+    address: 143.106.167.121
     parents: prokofiev
     hostgroups: computers, hassrvfs, buildd, stretch
   plummer:
-    address: 177.220.10.143
+    address: 143.106.167.122
     parents: prokofiev
     hostgroups: computers, porterbox, hassrvfs, stretch
   # }}}
@@ -1092,9 +1098,6 @@ hostgroups:
   armhf:
     alias: armhf
     private: 1
-  sparc:
-    alias: sparc
-    private: 1
 
   porterbox:
     alias: developer accessible porter machines
@@ -1138,8 +1141,6 @@ hostgroups:
     alias: Dell PowerEdge 1950 hosts
     private: 1
 
-  wheezy:
-    alias: Hosts running wheezy
   jessie:
     alias: Hosts running jessie
   stretch:
@@ -1183,9 +1184,6 @@ hostgroups:
   xinetd-hosts:
     alias: hosts providing services via xinetd
     private: 1
-  postgres91-hosts:
-    alias: hosts running postgres91
-    private: 1
   postgres94-hosts:
     alias: hosts running postgres94
     private: 1
@@ -1220,6 +1218,15 @@ hostgroups:
     alias: hosts with lots and lots of (kernel) processes
   crazymanyprocesses:
     alias: hosts with stupidly lots of processes
+  varnish-hosts:
+    alias: hosts running varnish
+    private: 1
+  haproxy-hosts:
+    alias: hosts running haproxy
+    private: 1
+  haproxy-https-host:
+    alias: "host providing https on the standard port via haproxy"
+    private: 1
 
   no-bacula:
     alias: hosts which are not being backed up with bacula
@@ -1475,6 +1482,38 @@ services:
     servicegroups: diskspace
     nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /storage/snapshot-farm-10"
     hosts: lw10
+
+  -
+    name: disk usage on nfs farm 1
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-1"
+    hosts: lw07
+  -
+    name: disk usage on nfs farm 2
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-2"
+    hosts: lw07
+  -
+    name: disk usage on nfs farm 3
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-3"
+    hosts: lw07
+  -
+    name: disk usage on nfs farm 4
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-4"
+    hosts: lw07
+  -
+    name: disk usage on nfs farm 09
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-09"
+    hosts: lw07
+  -
+    name: disk usage on nfs farm 10
+    servicegroups: diskspace
+    nrpe: "/usr/lib/nagios/plugins/check_disk 95 97 /auto.dsa/snapshot-10"
+    hosts: lw07
+
   -
     name: disk usage on /srv/morgue.debian.org/
     servicegroups: diskspace
@@ -1545,16 +1584,10 @@ services:
     name: free memory - percent
     nrpe: "/usr/lib/nagios/plugins/dsa-check-memory -m pct"
     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
-    excludehostgroups: jessie, stretch
   -
     name: process - getty
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:8 -c 1: -u root -C agetty -a /sbin/agetty"
-    hostgroups: jessie, stretch
+    hostgroups: computers
 
   -
     name: processes - zombies
@@ -1668,19 +1701,10 @@ services:
     name: puppetized firewall
     nrpe: "/usr/lib/nagios/plugins/dsa-check-file -w -f /etc/ferm/conf.d/defs.conf"
     hostgroups: computers
-  -
-    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: sparc, jessie, stretch
   -
     name: process - ulogd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u ulog -C ulogd -a '/usr/sbin/ulogd --daemon --uid ulog'"
-    hostgroups: jessie, stretch
-  -
-    name: unexpected process - ulogd
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C ulogd"
-    hostgroups: sparc
+    hostgroups: computers
   ####
   -
     name: process - samhain
@@ -1704,15 +1728,10 @@ services:
     excludehostgroups: brokensamhain
   # }}}
   # {{{ logging
-  -
-    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: jessie, stretch
   -
     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 -F'"
-    hostgroups: jessie, stretch
+    hostgroups: computers
 
   -
     name: remote logging on lotti
@@ -1793,7 +1812,7 @@ services:
     name: system time synced
     nrpe: "/usr/lib/nagios/plugins/dsa-check-timedatectl -s"
     hostgroups: computers
-    excludehostgroups: systemd-timesyncd, wheezy
+    excludehostgroups: systemd-timesyncd
     servicegroups: time
   -
     name: system time synced
@@ -1810,7 +1829,7 @@ 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: harris, smetana
+    excludehosts: harris
   ###
   -
     name: process - cron
@@ -1823,12 +1842,6 @@ services:
     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/bin/monit -d 300 -I -c /etc/monit/monitrc -s /var/lib/monit/monit.state'"
-    hostgroups: computers
-    excludehostgroups: alioth, jessie, stretch
-  ###
   -
     name: MQ connection on rainier
     servicegroups: MQ
@@ -1863,22 +1876,11 @@ services:
     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: 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: jessie, stretch
   -
     name: process - udevd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -p 1 -C systemd-udevd -a '/lib/systemd/systemd-udevd'"
-    hostgroups: jessie, stretch
+    hostgroups: computers
   ###
-  -
-    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
-    excludehostgroups: jessie, stretch
   -
     name: unexpected process - acpid
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C acpid"
@@ -1898,7 +1900,7 @@ services:
   -
     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: wheezy, jessie, stretch
+    hostgroups: computers
     excludehostgroups: alioth
   -
     name: process - stunnel4 - puppet-ekeyd is crazy
@@ -1965,18 +1967,11 @@ services:
     runfrom: handel
   # }}}
   # {{{ HW health/raid
-  -
-    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 /run/mdadm/monitor.pid --daemonise --scan'"
-    hostgroups: sw-raid
-    excludehostgroups: jessie, stretch
   -
     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 --scan'"
     hostgroups: sw-raid
-    excludehostgroups: wheezy
   -
     name: RAID - sw raid
     servicegroups: raid
@@ -2094,16 +2089,10 @@ services:
     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
-    excludehostgroups: jessie, stretch
   -
     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 --foreground=true'"
     hostgroups: heavy-exim, heavy-postfix
-    excludehostgroups: wheezy
   -
     name: unwanted process - clamav
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C clamd"
@@ -2116,18 +2105,11 @@ services:
     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
-    excludehostgroups: jessie, stretch
   -
     name: process - spamd - master
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd -d --pidfile=/var/run/spamd.pid --create-prefs --max-children 5 --helper-home-dir'"
     hostgroups: spamd
     excludehosts: picconi
-    excludehostgroups: wheezy
   -
     name: process - spamd - master
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C spamd -a '/usr/sbin/spamd -d --pidfile=/var/run/spamd.pid --create-prefs --max-children 20 --min-spare=5 --helper-home-dir'"
@@ -2156,16 +2138,10 @@ services:
     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
-    excludehostgroups: jessie, stretch
   -
     name: process - postgrey
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a 'postgrey --pidfile=/var/run/postgrey.pid --daemonize --unix=/var/run/postgrey/socket --retry-window=4 --auto-whitelist-clients=10 --exim'"
     hostgroups: heavy-exim
-    excludehostgroups: wheezy
   -
     name: process - postgrey
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgrey -a 'postgrey --pidfile=/var/run/postgrey.pid --daemonize --inet=127.0.0.1:60000'"
@@ -2364,15 +2340,52 @@ services:
   -
     name: network service - https cert
     check: dsa_check_cert!443
-    hostgroups: apache-https, https-service
+    hostgroups: apache-https, https-service, haproxy-https-host
     depends: network service - https
     check_interval: 60
   -
     name: unwanted network service - https
     check: dsa_check_port_closed!443
     hostgroups: apache2-hosts
-    excludehostgroups: apache-https
+    excludehostgroups: apache-https, haproxy-https-host
     check_interval: 60
+
+ ###
+  -
+    name: process - haproxy - master
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -a '/usr/sbin/haproxy-systemd-wrapper'"
+    hostgroups: haproxy-hosts
+  -
+    name: process - haproxy - worker
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:6 -c 1:15 -u haproxy -a '/usr/sbin/haproxy '"
+    hostgroups: haproxy-hosts
+    depends: process - haproxy - master
+  -
+    name: network service - https
+    check: check_https
+    hostgroups: haproxy-https-host
+    depends: "process - haproxy - master"
+    check_interval: 120
+
+  -
+    name: unwanted process - haproxy
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C haproxy"
+    hostgroups: computers
+    excludehostgroups: haproxy-hosts
+
+ ###
+  -
+    name: process - varnish
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -c 1:15 -u vcache -a '/usr/sbin/varnishd -j unix,user=vcache -F -a '"
+    hostgroups: varnish-hosts
+    excludehostgroups: jessie
+  -
+    name: unwanted process - varnish
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C varnishd"
+    hostgroups: computers
+    excludehostgroups: varnish-hosts
+
+
   # }}}
   # {{{ FTP
   -
@@ -2385,15 +2398,11 @@ services:
     name: unwanted process - postgresql
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C postgres"
     hostgroups: computers
-    excludehostgroups: postgres91-hosts, postgres94-hosts, postgres96-hosts
+    excludehostgroups: postgres94-hosts, postgres96-hosts
   -
     name: unwanted process - postgresql 9.0
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0 -C postgres -a '9.0/bin/postgres'"
     hostgroups: computers
-  -
-    name: process - postgresql91 - master
-    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/9.1/bin/postgres'"
-    hostgroups: postgres91-hosts
   -
     name: process - postgresql94 - master
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:10 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/9.4/bin/postgres'"
@@ -2820,7 +2829,7 @@ services:
   -
     name: system - all services running
     nrpe: "/usr/bin/sudo /usr/lib/nagios/plugins/dsa-check-systemd-services"
-    hostgroups: jessie, stretch
+    hostgroups: computers
   ###
   -
     name: process - slapd
@@ -2887,7 +2896,6 @@ services:
     name: puppet - agent check
     nrpe: "/usr/lib/nagios/plugins/dsa-check-statusfile /var/cache/dsa/nagios/puppet-agent"
     hostgroups: computers
-    excludehosts: moszumanska
     check_interval:  60
     retry_interval: 15
   # }}}