X-Git-Url: https://git.adam-barratt.org.uk/?a=blobdiff_plain;ds=sidebyside;f=nagios-master.cfg;h=6ababc2ff4fc7659d08fefa8da78d3c61947c8b7;hb=1dd1b851ae5287dba25ab9afdaa7349ccfd4255d;hp=6fbc709d10a8c287ca9f88003d04074e8c14c2db;hpb=87ded5a45f8720f8f7bcf555f1577575b0274a5a;p=mirror%2Fdsa-nagios.git diff --git a/nagios-master.cfg b/nagios-master.cfg index 6fbc709..6ababc2 100644 --- a/nagios-master.cfg +++ b/nagios-master.cfg @@ -17,12 +17,6 @@ # - puccini: mailgraph # - # - agnesi -# - casals -# - crest -# - elara -# - europa -# - kassia -# - kullervo # - lebrun # - murphy # - piatti @@ -118,6 +112,22 @@ servers: address: 193.201.200.129 parents: gw-HP-ftc hostgroups: routing-infrastructure + gw-xandros: + address: 142.46.212.33 + parents: gw-HP-ftc + hostgroups: routing-infrastructure + gw-nmmn: + address: 217.114.76.81 + parents: gw-HP-ftc + hostgroups: routing-infrastructure + gw-utwente: + address: 130.89.160.1 + parents: gw-HP-ftc + hostgroups: routing-infrastructure + gw-ughent: + address: 157.193.39.254 + parents: gw-HP-ftc + hostgroups: routing-infrastructure samosa: address: 192.25.206.57 @@ -179,7 +189,7 @@ servers: master: address: 70.103.162.29 parents: gw-brainfood - hostgroups: computers, general, apache2-hosts, bind9-hosts, heavy-exim + hostgroups: computers, general, apache2-hosts, bind9-hosts, heavy-exim, highload ries: address: 128.148.34.103 @@ -220,6 +230,10 @@ servers: address: 128.31.0.36 parents: gw-mit-csail hostgroups: computers, service, apache2-hosts, ftpd-hosts, rsyncd-hosts, sw-raid + casals: + address: 128.31.0.16 + parents: gw-mit-csail + hostgroups: deadslow klecker: address: 194.109.137.218 @@ -276,13 +290,41 @@ servers: puccini: address: 87.106.4.56 - parents: gw-frost + parents: gw-1und1 hostgroups: computers, service, apache2-hosts, bind9-hosts, postfix-hosts, heavy-postfix, amavis-hosts caballero: address: 193.201.200.200 parents: gw-blackcat - hostgroups: computers, buildd, bind9-hosts + hostgroups: computers, buildd, sw-raid + + elara: + address: 142.46.212.46 + parents: gw-xandros + hostgroups: deadslow + europa: + address: 142.46.212.46 + parents: gw-xandros + hostgroups: deadslow + + kullervo: + address: 217.114.76.82 + parents: gw-nmmn + hostgroups: deadslow + crest: + address: 217.114.76.83 + parents: gw-nmmn + hostgroups: deadslow + + kassia: + address: 130.89.175.54 + parents: gw-utwente + hostgroups: computers, service, postfix-hosts, apache2-hosts, ftpd-hosts, rsyncd-hosts, dl360 + + allegri: + address: 157.193.39.233 + parents: gw-ughent + hostgroups: computers, buildd, postfix-hosts, sw-raid ############################# # host groups @@ -298,6 +340,8 @@ hostgroups: alias: Internet routers and friends extinfo-icon_image: base/switch40.png extinfo-icon_image_alt: router + deadslow: + alias: Systems too slow to run any real checks porterbox: alias: developer accessible porter machines @@ -326,6 +370,9 @@ hostgroups: dl385: alias: HP DL385 hosts private: 1 + dl360: + alias: HP DL360 hosts + private: 1 sw-raid: alias: Hosts with Linux software raid private: 1 @@ -383,6 +430,17 @@ services: name: PING check: "check_ping!300.0,20%!600.0,40%" hostgroups: all + excludehostgroups: routing-infrastructure + normal_check_interval: 5 + max_check_attempts: 4 + retry_check_interval: 1 + - + name: PING + check: "check_ping!900.0,60%!1500.0,80%" + hostgroups: routing-infrastructure + normal_check_interval: 5 + max_check_attempts: 4 + retry_check_interval: 1 ############ Services ############ ### @@ -399,7 +457,7 @@ services: hostgroups: computers - name: disk usage on /boot - nrpe: "/usr/lib/nagios/plugins/check_disk 60 80 /boot" + nrpe: "/usr/lib/nagios/plugins/check_disk 75 85 /boot" hosts: sperger, rietz, steffani, penalosa, peri, albeniz, escher, goetz, mayer, mayr, paer - name: disk usage on /var @@ -408,11 +466,15 @@ services: - name: disk usage on /org nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org" - hosts: bartok, sperger, samosa, raff, lobos, villa, steffani, merkel, saens, pergolesi, verdi, puccini + hosts: bartok, sperger, samosa, raff, lobos, villa, steffani, saens, pergolesi, verdi, puccini + - + name: disk usage on /org + nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /org" + hosts: merkel - name: disk usage on /srv nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /srv" - hosts: agricola, arcadelt, argento + hosts: agricola, arcadelt, argento, allegri - name: disk usage on /org/scratch2 nrpe: "/usr/lib/nagios/plugins/check_disk 80 90 /org/scratch2" @@ -432,7 +494,11 @@ services: - name: disk usage on /home nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /home" - hosts: gluck, raptor, escher, voltaire + hosts: raptor, escher, voltaire + - + name: disk usage on /home + nrpe: "/usr/lib/nagios/plugins/check_disk 90 95 /home" + hosts: gluck - name: disk usage on /chroot nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /chroot" @@ -444,7 +510,7 @@ services: - name: disk usage on /x nrpe: "/usr/lib/nagios/plugins/check_disk 75 90 /x" - hosts: voltaire + hosts: caballero ############ All Computers ############ #### @@ -464,13 +530,18 @@ services: #### - name: load - nrpe: "/usr/lib/nagios/plugins/check_load -w 12,10,8 -c 22,18,14" + nrpe: "/usr/lib/nagios/plugins/check_load -w 30,28,26 -c 50,45,50" hostgroups: computers excludehostgroups: highload - name: load - nrpe: "/usr/lib/nagios/plugins/check_load -w 100,80,60 -c 200,160,140" + nrpe: "/usr/lib/nagios/plugins/check_load -w 140,120,100 -c 240,220,200" hostgroups: highload + excludehosts: rietz + - + name: load + nrpe: "/usr/lib/nagios/plugins/check_load -w 200,200,200 -c 350,350,350" + hosts: rietz #### - name: processes - zombies @@ -508,6 +579,13 @@ services: depends: process - sshd normal_check_interval: 60 notification_interval: 60 + + - + name: "network service - sshd" + check: check_ssh + hostgroups: deadslow + normal_check_interval: 180 + notification_interval: 180 #### - name: network service - nrpe @@ -710,14 +788,8 @@ services: depends: process - weightd - master - name: process - weightd - child - nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:10 -c 1: -u polw -C policyd-weight -a 'policyd-weight (child)'" + nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:50 -c 1: -u polw -C policyd-weight -a 'policyd-weight (child)'" hostgroups: heavy-postfix - excludehosts: liszt - depends: process - weightd - master - - - name: process - weightd - child - nrpe: "/usr/lib/nagios/plugins/check_procs -w 2:40 -c 1: -u polw -C policyd-weight -a 'policyd-weight (child)'" - hosts: liszt depends: process - weightd - master # - @@ -777,7 +849,7 @@ services: - name: process - postfix - smtpd - nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:10 -c 0:15 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'" + nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:20 -c 0:50 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'" hostgroups: postfix-hosts excludehosts: liszt depends: process - postfix - master @@ -790,7 +862,7 @@ services: - name: process - postfix - smtpd - nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:100 -c 0:150 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'" + nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:150 -c 0:200 -u postfix -C smtpd -a 'smtpd -n smtp -t inet -u -c'" hosts: liszt depends: process - postfix - master - @@ -802,12 +874,23 @@ services: ### - name: network service - smtp - check: check_smtp + check: dsa_check_smtp hostgroups: postfix-hosts + excludehosts: verdi, kassia, allegri + depends: process - postfix - master + - + name: network service - smtp - port 2025 + check: dsa_check_smtp_port!2025 + hosts: verdi, kassia + depends: process - postfix - master + - + name: network service - smtp - port 8080 + check: dsa_check_smtp_port!8080 + hosts: allegri depends: process - postfix - master - name: network service - smtp - check: check_smtp + check: dsa_check_smtp hostgroups: computers excludehostgroups: postfix-hosts depends: process - exim @@ -823,19 +906,19 @@ services: name: process - irqbalance nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C irqbalance -a '/usr/sbin/irqbalance'" hostgroups: computers - excludehosts: arcadelt, agricola, argento, penalosa, peri, escher, bruckner + excludehosts: arcadelt, agricola, argento, penalosa, peri, escher, bruckner, allegri #### ### - - - name: unwanted process - system-tools-backends - nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C system-tools-ba" - hostgroups: computers - - - name: unwanted process - dbus-daemon - nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C dbus-daemon" - hostgroups: computers + #- + # name: unwanted process - system-tools-backends + # nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C system-tools-ba" + # hostgroups: computers + #- + # name: unwanted process - dbus-daemon + # nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C dbus-daemon" + # hostgroups: computers - name: unwanted process - gkrellmd nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C gkrellmd" @@ -856,7 +939,7 @@ services: name: unwanted process - snmpd nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C snmpd" hostgroups: computers - excludehosts: lobos, villa, caballero + excludehosts: lobos, villa ############ Processes/Services that only run on some computers ############ @@ -886,11 +969,11 @@ services: - name: process - cpqarrayd nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u root -C cpqarrayd -a '/usr/sbin/cpqarrayd'" - hostgroups: dl385, dl380 + hostgroups: dl385, dl380, dl360 - name: RAID - arrayprobe nrpe: "sudo /usr/bin/arrayprobe" - hostgroups: dl385, dl380 + hostgroups: dl385, dl380, dl360 ### - @@ -901,12 +984,12 @@ services: - name: process - udevd nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C udevd -a 'udevd'" - hosts: sperger, ries, steffani, merkel, spohr, peri, penalosa, albeniz, escher, verdi, liszt + hosts: sperger, ries, steffani, merkel, spohr, peri, penalosa, albeniz, escher, verdi, liszt, kassia, agricola, arcadelt, argento, allegri ### - name: process - acpid nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C acpid -a '/usr/sbin/acpid -c /etc/acpi/events -s /var/run/acpid.socket'" - hosts: ries, steffani, liszt + hosts: ries, steffani, liszt, kassia ### - @@ -1026,7 +1109,7 @@ services: hostgroups: ftpd-hosts - name: process - vsftp - instance - nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:30 -c 0:50 -u ftp -C vsftpd -a 'vsftpd: '" + nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:80 -c 0:100 -u ftp -C vsftpd -a 'vsftpd: '" hostgroups: ftpd-hosts - name: network service - ftp @@ -1066,7 +1149,7 @@ services: - name: process - snmpd nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1:1 -u snmp -C snmpd -a '/usr/sbin/snmpd -Lsd -Lf /dev/null -u snmp -I -smux -p /var/run/snmpd.pid 127.0.0.1'" - hosts: lobos, villa, caballero + hosts: lobos, villa ### - @@ -1090,3 +1173,8 @@ services: name: process - dhcpd nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C dhcpd3 -a '/usr/sbin/dhcpd3 -q eth0'" hosts: paer + ### + - + name: process - monit + nrpe: "/usr/lib/nagios/plugins/check_procs -w 1:1 -c 1: -u root -C monit -a '/usr/sbin/monit -d 30 -c /etc/monit/monitrc -s /var/lib/monit/monit.state'" + hosts: villa, lobos, steffani, kassia, master