X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;f=config%2Fnagios-master.cfg;h=f54c7eb4a9af092fc2163dfec0e41e0b677c1ee7;hb=36b442bd9996a92442e83dd44b42c7ff155c94b5;hp=8a86a4fee7e8fa1601d984f5bebce15d6df3d9ef;hpb=411a177c41f3b3fe9fe64cbb7ae6433255d89712;p=mirror%2Fdsa-nagios.git diff --git a/config/nagios-master.cfg b/config/nagios-master.cfg index 8a86a4f..f54c7eb 100644 --- a/config/nagios-master.cfg +++ b/config/nagios-master.cfg @@ -68,7 +68,7 @@ servers: parents: gw-ubc hostgroups: routing-infrastructure gw-dg-i.net: - address: 93.94.130.190 + address: 93.94.132.102 parents: gw-ubc hostgroups: routing-infrastructure contacts: bzed @@ -141,7 +141,7 @@ servers: parents: gw-ubc hostgroups: routing-infrastructure gw-cst: - address: 213.157.2.210 + address: 213.157.2.23 parents: gw-ubc hostgroups: routing-infrastructure gw-isc: @@ -197,12 +197,12 @@ servers: peri: address: 192.25.206.15 parents: gw-HP-ftc - hostgroups: computers, buildd, sw-raid, hasbootfs, single-cpu, ulogd-hosts + hostgroups: computers, buildd, sw-raid, hasbootfs, ulogd-hosts contacts: dannf penalosa: address: 192.25.206.68 parents: gw-HP-ftc - hostgroups: computers, buildd, hasbootfs, single-cpu, ulogd-hosts + hostgroups: computers, buildd, hasbootfs, ulogd-hosts contacts: dannf mundy: address: 192.25.206.62 @@ -211,7 +211,7 @@ servers: paer: address: 192.25.206.11 parents: gw-HP-ftc - hostgroups: computers, porterbox, hasbootfs, hashomefs, single-cpu + hostgroups: computers, porterbox, hasbootfs, hashomefs merulo: address: 192.25.206.58 parents: gw-HP-ftc @@ -234,23 +234,22 @@ servers: agricola: address: 82.195.75.86 parents: gw-man-da - hostgroups: computers, porterbox, sw-raid, single-cpu, hassrvfs + hostgroups: computers, porterbox, sw-raid, hassrvfs contacts: bzed arcadelt: address: 82.195.75.87 parents: gw-man-da - hostgroups: computers, buildd, sw-raid, single-cpu, hassrvfs, ulogd-hosts + hostgroups: computers, buildd, sw-raid, hassrvfs, ulogd-hosts 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 + heininen: + address: 82.195.75.98 + parents: gw-man-da + hostgroups: computers, ulogd, acpid unger: address: 82.195.75.102 parents: gw-man-da @@ -270,11 +269,7 @@ servers: kaufmann: address: 82.195.75.107 parents: unger - hostgroups: computers, service, hasbootfs, apache2-hosts, single-cpu, rsyncd-hosts, kvmdomains, xinetd-hosts - byrd: - address: 82.195.75.101 - parents: unger - hostgroups: computers, service, hasbootfs, hassrvfs, postgres84-hosts, heavy-exim, kvmdomains + hostgroups: computers, service, hasbootfs, apache2-hosts, rsyncd-hosts, kvmdomains, xinetd-hosts ball: address: 82.195.75.70 parents: gw-man-da @@ -282,7 +277,7 @@ servers: purcell: address: 82.195.75.69 parents: gw-man-da - hostgroups: computers, buildd, single-cpu, ulogd-hosts + hostgroups: computers, buildd, ulogd-hosts master: address: 70.103.162.29 @@ -310,7 +305,7 @@ servers: rietz: address: 140.211.166.43 parents: gw-osuosl - hostgroups: computers, no-udev, service, apache2-hosts, bind9-hosts, rsyncd-hosts, dl385, heavy-exim, highload, hasbootfs, etch, xinetd-hosts + hostgroups: computers, service, rsyncd-hosts, dl385, hassrvfs, acpid-hosts, xinetd-hosts rietz2: address: 140.211.166.44 parents: rietz @@ -344,7 +339,7 @@ servers: #mahler: # address: 128.31.0.46 # parents: gw-mit-csail - # hostgroups: computers, single-cpu, porterbox, hasorgfs + # hostgroups: computers, porterbox, hasorgfs rore: address: 128.31.0.49 parents: gw-mit-csail @@ -371,7 +366,7 @@ servers: argento: address: 93.94.130.160 parents: gw-dg-i.net - hostgroups: computers, buildd, sw-raid, single-cpu, hassrvfs, ulogd-hosts + hostgroups: computers, buildd, sw-raid, hassrvfs, ulogd-hosts contacts: bzed widor: @@ -389,13 +384,13 @@ servers: address: 193.62.202.27 parents: gw-sanger # SMP kernel doesn't run stable - hostgroups: computers, porterbox, sw-raid, single-cpu, hasbootfs, ntpsuckers + hostgroups: computers, porterbox, sw-raid, hasbootfs, ntpsuckers contacts: tjrc1 goetz: address: 193.62.202.26 parents: gw-sanger # SMP kernel doesn't run stable - hostgroups: computers, buildd, sw-raid, single-cpu, hasbootfs, ntpsuckers, hassrvfs, ulogd-hosts + hostgroups: computers, buildd, sw-raid, hasbootfs, ntpsuckers, hassrvfs, ulogd-hosts contacts: tjrc1 sibelius: address: 193.62.202.28 @@ -440,7 +435,7 @@ servers: geo3: address: 195.20.242.125 parents: schumann - hostgroups: computers, service, hasbootfs, single-cpu, bind9-hosts, kvmdomains, ulogd-hosts + hostgroups: computers, service, hasbootfs, bind9-hosts, kvmdomains, ulogd-hosts soler: address: 195.20.242.126 parents: schumann @@ -454,7 +449,7 @@ servers: pescetti: address: 217.114.76.85 parents: gw-nmmn - hostgroups: computers, porterbox, single-cpu, smart, hasorgfs + hostgroups: computers, porterbox, smart, hasorgfs kassia: address: 130.89.149.224 @@ -476,7 +471,7 @@ servers: ancina: address: 157.193.39.13 parents: gw-ghent - hostgroups: computers, buildd, single-cpu, hassrvfs, hasbootfs, incomingmailrelayed2025, xinetd-hosts, ulogd-hosts + hostgroups: computers, buildd, hassrvfs, hasbootfs, incomingmailrelayed2025, xinetd-hosts, ulogd-hosts contacts: luk agnesi: @@ -503,7 +498,7 @@ servers: hostgroups: computers, bl460, acpid-hosts, ulogd-hosts tchaikovsky: address: 206.12.19.118 - hostgroups: computers, general, apache2-hosts, hasbootfs, kvmdomains, single-cpu, ulogd-hosts + hostgroups: computers, general, apache2-hosts, hasbootfs, kvmdomains, ulogd-hosts wolkenstein: address: 206.12.19.116 parents: dijkstra @@ -515,7 +510,7 @@ servers: geo2: address: 206.12.19.113 parents: dijkstra - hostgroups: computers, service, hasbootfs, single-cpu, bind9-hosts, kvmdomains, ulogd-hosts + hostgroups: computers, service, hasbootfs, bind9-hosts, kvmdomains, ulogd-hosts duarte: address: 206.12.19.114 parents: dijkstra @@ -523,7 +518,7 @@ servers: valente: address: 206.12.19.112 parents: dijkstra - hostgroups: computers, hasbootfs, single-cpu, hassrvfs, postgres83-hosts, ftpd-hosts, rsyncd-hosts, kvmdomains, heavy-exim, xinetd-hosts + hostgroups: computers, hasbootfs, hassrvfs, postgres83-hosts, ftpd-hosts, rsyncd-hosts, kvmdomains, heavy-exim, xinetd-hosts bellini: address: 206.12.19.7 parents: gw-ubc @@ -531,7 +526,7 @@ servers: morricone: address: 206.12.19.9 parents: tchaikovsky - hostgroups: computers, hasbootfs, hassrvfs, aacraid, postgres83-hosts, ftpd-hosts, rsyncd-hosts, heavy-exim, xinetd-hosts + hostgroups: computers, hasbootfs, hassrvfs, aacraid, postgres83-hosts, ftpd-hosts, rsyncd-hosts, heavy-exim, xinetd-hosts, apache2-hosts stabile: address: 206.12.19.13 parents: gw-ubc @@ -571,16 +566,20 @@ servers: gabrielli: address: 206.12.19.17 parents: tchaikovsky - hostgroups: computers, single-cpu + hostgroups: computers corelli: address: 206.12.19.16 parents: tchaikovsky - hostgroups: computers, single-cpu, buildd, ulogd-hosts + hostgroups: computers, buildd, ulogd-hosts + lucatelli: + address: 206.12.19.15 + parents: tchaikovsky + hostgroups: computers, buildd, ulogd-hosts lebrun: address: 193.198.184.10 parents: gw-carnet - hostgroups: computers, buildd, sw-raid, hasorgfs, ulogd-hosts + hostgroups: computers, buildd, sw-raid, hasorgfs schroeder: address: 193.198.184.11 parents: gw-carnet @@ -614,7 +613,7 @@ servers: goedel: address: 213.188.99.214 parents: gw-cst - hostgroups: computers, single-cpu, hasbootfs, ntpsuckers, ulogd-hosts + hostgroups: computers, hasbootfs, ntpsuckers, ulogd-hosts schein: address: 149.20.20.6 @@ -628,7 +627,7 @@ servers: pettersson: address: 130.239.18.123 parents: gw-aca - hostgroups: computers, hasbootfs, aacraid, nfs-client + hostgroups: computers, hasbootfs, aacraid, nfs-client, acpid-hosts lafayette: address: 147.215.7.160 @@ -689,11 +688,15 @@ servers: beethoven: address: 86.59.118.146 parents: gw-sil - hostgroups: computers, hasbootfs, aacraid, ulogd-hosts + hostgroups: computers, hasbootfs, aacraid, ulogd-hosts, nfs-server merikanto: address: 86.59.118.147 parents: gw-sil - hostgroups: computers, dl360, acpid-hosts, spamd, heavy-exim + hostgroups: computers, dl360, acpid-hosts, spamd, heavy-exim, nfs-client, rsyncd-hosts, xinetd-hosts + merikanto2: + address: 86.59.118.150 + parents: merikanto + hostgroups: secondary-IPs englund: address: 86.59.118.148 parents: gw-sil @@ -774,9 +777,6 @@ hostgroups: mptraid: alias: Hosts with LSI Logic Fusion-MPT private: 1 - single-cpu: - alias: Hosts with only one CPU - private: 1 etch: alias: Hosts running etch @@ -947,7 +947,7 @@ services: - name: disk usage - all servicegroups: diskspace - nrpe: "/usr/lib/nagios/plugins/check_disk -X devpts -X proc -X linprocfs -X devfs -X fdescfs -X sysfs 90 95" + nrpe: "/usr/lib/nagios/plugins/check_disk -X devpts -X proc -X linprocfs -X devfs -X fdescfs -X sysfs -X nfs 90 95" hostgroups: computers - name: disk usage on / @@ -1019,7 +1019,13 @@ services: name: disk usage on /var/lib/postgresql servicegroups: diskspace nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /var/lib/postgresql" - hosts: samosa, stabile, sibelius + hosts: samosa, stabile, sibelius, busoni + + - + name: disk usage on /var/log + servicegroups: diskspace + nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /var/log" + hosts: busoni - name: disk usage on /srv/morgue.debian.org @@ -1030,7 +1036,7 @@ services: name: disk usage on /srv/mirrors servicegroups: diskspace nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /srv/mirrors" - hosts: stabile + hosts: stabile, beethoven, sibelius - name: disk usage on /srv/ftp-master.debian.org servicegroups: diskspace @@ -1042,24 +1048,39 @@ services: nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /srv/snapshot.debian.org" hosts: stabile,sibelius - - name: disk usage on /srv/farm-snapshot/farm-0123 + name: disk usage on /srv/farm-snapshot/farm-01 servicegroups: diskspace - nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-0123" + nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-01" hosts: sibelius - - name: disk usage on /srv/farm-snapshot/farm-4567 + name: disk usage on /srv/farm-snapshot/farm-23 servicegroups: diskspace - nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-4567" + nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-23" hosts: sibelius - - name: disk usage on /srv/farm-snapshot/farm-89ab + name: disk usage on /srv/farm-snapshot/farm-45 servicegroups: diskspace - nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-89ab" + nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-45" hosts: sibelius - - name: disk usage on /srv/farm-snapshot/farm-cdef + name: disk usage on /srv/farm-snapshot/farm-67 servicegroups: diskspace - nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-cdef" + nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-67" + hosts: sibelius + - + name: disk usage on /srv/farm-snapshot/farm-89 + servicegroups: diskspace + nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-89" + hosts: sibelius + - + name: disk usage on /srv/farm-snapshot/farm-abf + servicegroups: diskspace + nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-abf" + hosts: sibelius + - + name: disk usage on /srv/farm-snapshot/farm-cde + servicegroups: diskspace + nrpe: "/usr/lib/nagios/plugins/check_disk 85 95 /srv/farm-snapshot/farm-cde" hosts: sibelius ############ All Computers ############ @@ -1132,11 +1153,11 @@ services: 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: uptime check + nrpe: "/usr/lib/nagios/plugins/dsa-check-uptime" + hostgroups: computers #### - name: processes - zombies @@ -1257,6 +1278,12 @@ 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: computers + excludehostgroups: squeeze + + - + 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: squeeze ### MAIL STUFF ### @@ -1265,17 +1292,17 @@ services: 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, byrd + excludehosts: master, busoni, merkel - name: process - exim - total nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:50 -c 1: -C exim4" hostgroups: computers excludehostgroups: postfix-hosts - excludehosts: master, rietz, merkel, byrd + excludehosts: master, busoni, merkel - name: process - exim nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:300 -c 1:500 -C exim4 -a '/usr/sbin/exim4'" - hosts: master, rietz, merkel, byrd + hosts: master, busoni, merkel ### - name: process - clamav - clamd @@ -1334,7 +1361,7 @@ services: nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C spamd" hostgroups: computers excludehostgroups: spamd - excludehosts: liszt, rietz + excludehosts: liszt, busoni ### #- @@ -1506,15 +1533,10 @@ services: nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u daemon -C uptimed -a '/usr/sbin/uptimed'" hostgroups: computers ### - - - 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 - excludehostgroups: single-cpu, freebsd - name: unwanted process - irqbalance nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C irqbalance" - hostgroups: single-cpu, freebsd + hostgroups: computers #### ### @@ -1544,7 +1566,7 @@ services: name: unwanted process - inetd nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C inetd" hostgroups: computers - excludehosts: rietz, grieg + excludehosts: grieg - name: unwanted process - snmpd nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C snmpd" @@ -1578,14 +1600,8 @@ services: ### - 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 -c /etc/monit/monitrc -s /var/lib/monit/monit.state'" + 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: computers - excludehostgroups: squeeze - - - name: process - monit - nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C monit -a '/usr/sbin/monit -c /etc/monit/monitrc -s /var/lib/monit/monit.state'" - hostgroups: squeeze - ### - name: process - cpqarrayd servicegroups: raid @@ -1713,23 +1729,27 @@ services: hosts: draghi depends: process - xinetd ### - - - name: network service - rsync - check: check_tcp!873 - hostgroups: rsyncd-hosts - depends: process - xinetd - +# - +# name: network service - rsync +# check: check_tcp!873 +# hostgroups: rsyncd-hosts +# depends: process - xinetd +# - name: network service - rsync check: check_tcp!873 hosts: rietz2 depends: rietz:process - xinetd - - name: network service - rsync check: check_tcp!873 hosts: kassia-sec depends: kassia:process - xinetd + - + name: network service - rsync + check: check_tcp!873 + hosts: merikanto2 + depends: merikanto:process - xinetd ### - @@ -1772,7 +1792,7 @@ services: - name: network service - https check: check_https - hosts: ries, chopin, draghi, liszt, spohr, widor, rietz, tchaikovsky, grieg + hosts: ries, chopin, draghi, liszt, spohr, widor, tchaikovsky, grieg depends: "process - apache2 - master" normal_check_interval: 120 - @@ -1785,8 +1805,8 @@ services: # draghi db.debian.org # merkel2 nm.debian.org # liszt lists.debian.org - # cimarosa buildd.debian.org - hosts: ries, chopin, spohr, tchaikovsky, draghi, merkel2, liszt, widor, rietz, grieg + # grieg buildd.debian.org + hosts: ries, chopin, spohr, tchaikovsky, draghi, merkel2, liszt, widor, grieg depends: network service - https normal_check_interval: 60 @@ -1892,6 +1912,11 @@ services: name: process - postresql84 - dak-dev master nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u postgres -C postgres -a '/usr/lib/postgresql/8.4/bin/postgres -D /var/lib/postgresql/8.4/dak-dev -c config_file=/etc/postgresql/8.4/dak-dev/postgresql.conf'" hosts: ries + # backups + - + name: postgresql backups + nrpe: "/usr/bin/sudo -u debbackup /usr/lib/nagios/plugins/dsa-check-backuppg" + hosts: beethoven #### - name: process - mysql - master @@ -1937,7 +1962,7 @@ services: - name: process - buildd servicegroups: buildd - nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:2 -u buildd -C buildd -a '/usr/bin/buildd'" + nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:5 -u buildd -C buildd -a '/usr/bin/buildd'" hostgroups: buildd contacts: luk, pkern, HE @@ -1994,19 +2019,26 @@ services: ##### - name: mirror sync - bugs - check: "dsa_check_mirrorsync!bugs.debian.org!project/trace/bugs-master.debian.org" + check: "dsa_check_mirrorsync_skew!bugs.debian.org!project/trace/bugs-master.debian.org!120:600" hosts: global - name: mirror sync - security - check: "dsa_check_mirrorsync!security-nagios.debian.org!project/trace/security-master.debian.org" + check: "dsa_check_mirrorsync_skew!security-nagios.debian.org!project/trace/security-master.debian.org!150:3600" hosts: global - name: mirror sync - volatile - check: "dsa_check_mirrorsync!volatile.debian.org!debian-volatile/project/trace/volatile-master.debian.org" + check: "dsa_check_mirrorsync_skew!volatile.debian.org!debian-volatile/project/trace/volatile-master.debian.org!150:3600" hosts: global - name: mirror sync - packages - check: "dsa_check_mirrorsync!packages.debian.org!Pics/.trace" + check: "dsa_check_mirrorsync_skew!packages.debian.org!Pics/.trace!3600:57600" + hosts: global + normal_check_interval: 15 + max_check_attempts: 5 + retry_check_interval: 5 + - + name: mirror sync - snapshot + check: "dsa_check_mirrorsync_skew!snapshot.debian.org!project/trace/snapshot-master.debian.org!1800:28800" hosts: global normal_check_interval: 15 max_check_attempts: 5