X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=config%2Fnagios-master.cfg;h=8f46b7a44dc443073aa52d0dd4c6db053e91a5d7;hb=637e5f965db829fc40539b1786bc700e1677bfd1;hp=a432e4e729a08aa8841efcc3b3a67075efe12a5f;hpb=e313f7d1d139ad3de5ae8d99a49f33244fc3aacb;p=mirror%2Fdsa-nagios.git

diff --git a/config/nagios-master.cfg b/config/nagios-master.cfg
index a432e4e..8f46b7a 100644
--- a/config/nagios-master.cfg
+++ b/config/nagios-master.cfg
@@ -85,10 +85,6 @@ servers:
     address: 72.52.94.70
     parents: gw-ubcece
     hostgroups: layer3-infrastructure
-  gw-karlsruhe:
-    address: 129.143.59.214
-    parents: gw-ubcece
-    hostgroups: layer3-infrastructure
   gw-leaseweb:
     address: 185.17.185.190
     parents: gw-ubcece
@@ -143,7 +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:
@@ -409,9 +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
@@ -447,7 +441,7 @@ servers:
   x86-bm-01:
     address: 5.153.231.32
     parents: ganeti-bytemark
-    hostgroups: computers, kvmdomains, stretch, no-bacula, systemd-timesyncd
+    hostgroups: computers, pybuildd, hassrvfs, kvmdomains, stretch, systemd-timesyncd
   tate:
     address: 5.153.231.33
     parents: ganeti-bytemark
@@ -597,7 +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
@@ -643,7 +637,7 @@ servers:
   x86-grnet-01:
     address: 194.177.211.203
     parents: ganeti-grnet
-    hostgroups: computers, buildd, hassrvfs, kvmdomains, stretch, systemd-timesyncd
+    hostgroups: computers, pybuildd, hassrvfs, kvmdomains, stretch, systemd-timesyncd
   vittoria:
     address: 194.177.211.205
     parents: ganeti-grnet
@@ -687,27 +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
@@ -717,13 +716,6 @@ servers:
     parents: gw-leaseweb
     hostgroups: computers, service, stretch, dl180
   # }}}
-  # {{{ gw-karlsruhe
-  zemlinsky:
-    address: 129.143.160.6
-    parents: gw-karlsruhe
-    hostgroups: computers, buildd, stretch
-    contacts: pkern
-  # }}}
   # {{{ gw-manda
   czerny:
     address: 82.195.75.109
@@ -818,7 +810,7 @@ servers:
   zani:
     address: 148.100.88.22
     parents: gw-marist
-    hostgroups: computers, buildd, hassrvfs, stretch, incomingmailrelayed
+    hostgroups: computers, pybuildd, hassrvfs, stretch, incomingmailrelayed
   # }}}
   # {{{ gw-osuosl
   byrd:
@@ -842,20 +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
+    hostgroups: computers, postgres94-hosts, service, apache2-hosts, sw-raid, jessie, rsyncd-hosts, hasvarlogfs, multipath-hosts, nfs-server, varnish-hosts
     contacts: tjrc1, dave
   smetana:
     address: 193.62.202.29
@@ -995,11 +987,19 @@ servers:
   godard:
     address: 209.87.16.44
     parents: ubc-gateway
-    hostgroups: computers, service, kvmdomains, stretch, hassrvfs, apache2-hosts, apache-https, systemd-timesyncd, postfix-hosts, postgres96-hosts, manyprocesses
+    hostgroups: computers, service, kvmdomains, stretch, hassrvfs, apache2-hosts, apache-https, systemd-timesyncd, postfix-hosts, postgres96-hosts, crazymanyprocesses
   debussy:
     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:
@@ -1025,19 +1025,19 @@ 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
   # }}}
@@ -1113,6 +1113,8 @@ hostgroups:
     alias: machines running services
   buildd:
     alias: buildd systems
+  pybuildd:
+    alias: buildd systems running pybuildd
   general:
     alias: general purpose developer accessible machines
 
@@ -1192,9 +1194,6 @@ 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
@@ -1227,6 +1226,17 @@ hostgroups:
     private: 1
   manyprocesses:
     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
@@ -1482,6 +1492,38 @@ 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
@@ -1535,11 +1577,15 @@ services:
     name: processes - total
     nrpe: "/usr/lib/nagios/plugins/check_procs 620 700"
     hostgroups: computers
-    excludehostgroups: manyprocesses
+    excludehostgroups: manyprocesses, crazymanyprocesses
   -
     name: processes - total
     hostgroups: manyprocesses
     nrpe: "/usr/lib/nagios/plugins/check_procs 1500 1700"
+  -
+    name: processes - total
+    hostgroups: crazymanyprocesses
+    nrpe: "/usr/lib/nagios/plugins/check_procs 15000 25000"
   -
     name: free memory - mb
     nrpe: "/usr/lib/nagios/plugins/dsa-check-memory -m mb"
@@ -1599,7 +1645,7 @@ services:
     remotecheck: "/usr/lib/nagios/plugins/dsa-check-bacula $HOSTNAME$.debian.org"
     runfrom: dinis
     hostgroups: computers
-    excludehostgroups: buildd, porterbox, no-bacula
+    excludehostgroups: buildd, pybuildd, porterbox, no-bacula
     check_interval:  60
     retry_interval: 15
   -
@@ -1608,7 +1654,7 @@ services:
     remotecheck: "/usr/lib/nagios/plugins/dsa-check-bacula -w 1080 -c 1560 $HOSTNAME$.debian.org F"
     runfrom: dinis
     hostgroups: computers
-    excludehostgroups: buildd, porterbox, no-bacula
+    excludehostgroups: buildd, pybuildd, porterbox, no-bacula
     check_interval:  60
     retry_interval: 15
   -
@@ -1671,11 +1717,6 @@ 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'"
@@ -1877,11 +1918,6 @@ services:
     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
   ###
-  -
-    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"
@@ -1961,11 +1997,6 @@ services:
     name: "sso CRL"
     nrpe: "if [ -e /var/lib/dsa/sso/ca.crl ]; then /usr/lib/nagios/plugins/dsa-check-crl-expire -w 129600 -c 86400 /var/lib/dsa/sso/ca.crl; else echo 'No sso/ca.crl on this host.'; fi"
     hostgroups: computers
-  -
-    name: SSL certs - puppet
-    hosts: global
-    remotecheck: "/usr/lib/nagios/plugins/dsa-check-cert-expire-dir /etc/puppet/modules/ssl/files/servicecerts"
-    runfrom: handel
   -
     name: SSL certs - LE
     hosts: global
@@ -2102,16 +2133,10 @@ 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"
@@ -2124,18 +2149,11 @@ 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'"
@@ -2164,16 +2182,10 @@ 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'"
@@ -2372,15 +2384,52 @@ 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
   -
@@ -2393,15 +2442,11 @@ 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'"
@@ -2422,19 +2467,20 @@ services:
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:5 -u buildd -C buildd -a '/usr/bin/buildd'"
     hostgroups: buildd
     contact_groups: buildd
+  -
+    name: process - buildd
+    servicegroups: buildd
+    nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:5 -u buildd -C python3 -a 'buildd.py'"
+    hostgroups: pybuildd
+    contact_groups: buildd
   -
     name: processes - zombie schroot
     nrpe: "(/usr/lib/nagios/plugins/check_procs -a schroot -s Zs -c 0 > /dev/null || /usr/lib/nagios/plugins/check_procs -a schroot -s Zs -c 0) && /usr/lib/nagios/plugins/check_procs -a schroot -s ZNs -c 0"
-    hostgroups: buildd
+    hostgroups: buildd, pybuildd
     contact_groups: +buildd
     check_interval: 5
     max_check_attempts: 24
     retry_interval: 5
-  -
-    name: processes - lvcreate
-    nrpe: "/usr/lib/nagios/plugins/check_procs -m 'ELAPSED' -c 500 -C lvcreate -u root -a 'lvcreate'"
-    hostgroups: buildd
-    contact_groups: +buildd
   # }}}
   # {{{ NFS Stuff
   -
@@ -2894,7 +2940,6 @@ 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
   # }}}