the bm blades have lots of new kernel processes

[mirror/dsa-nagios.git] / config / nagios-master.cfg

diff --git a/config/nagios-master.cfg b/config/nagios-master.cfg
index dbc7b07..705f250 100644 (file)
--- a/config/nagios-master.cfg
+++ b/config/nagios-master.cfg
@@ -271,65 +271,65 @@ servers:
   fasolo:
     address: 138.16.160.17
     parents: gw-brown
-    hostgroups: computers, service, apache2-hosts, apache-https, dl380, rsyncd-systemd-hosts, jessie, hassrvfs, postgres94-hosts
+    hostgroups: computers, service, apache2-hosts, apache-https, dl380, rsyncd-systemd-hosts, jessie, hassrvfs, postgres94-hosts, manyprocesses
   # }}}
   # {{{ gw-bytemark
   bm-bl1:
     address: 5.153.231.241
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl2:
     address: 5.153.231.242
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl3:
     address: 5.153.231.243
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl4:
     address: 5.153.231.244
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl5:
     address: 5.153.231.245
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl6:
     address: 5.153.231.246
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl7:
     address: 5.153.231.247
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl8:
     address: 5.153.231.248
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl9:
     address: 5.153.231.249
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl10:
     address: 5.153.231.250
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl11:
     address: 5.153.231.251
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl12:
     address: 5.153.231.252
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl13:
     address: 5.153.231.253
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
   bm-bl14:
     address: 5.153.231.254
     parents: gw-bytemark
-    hostgroups: computers, bm-bl, service, jessie, multipath-hosts
+    hostgroups: computers, bm-bl, service, stretch, multipath-hosts, manyprocesses
 
   milanollo:
     address: 5.153.231.2
@@ -376,7 +376,7 @@ servers:
   coccia:
     address: 5.153.231.11
     parents: ganeti-bytemark
-    hostgroups: computers, hassrvfs, kvmdomains, jessie, autofs, nfs-client, uploadqueue, apache-https, apache2-hosts, systemd-timesyncd
+    hostgroups: computers, hassrvfs, kvmdomains, stretch, autofs, nfs-client, uploadqueue, apache-https, apache2-hosts, systemd-timesyncd
   backuphost:
     address: 5.153.231.12
     parents: ganeti-bytemark
@@ -503,7 +503,7 @@ servers:
   casulana:
     address: 5.153.231.41
     parents: gw-bytemark
-    hostgroups: computers, service, stretch, hassrvfs, dl380
+    hostgroups: computers, service, stretch, hassrvfs, dl380, manyprocesses
   # }}}
   # {{{ gw-c3sl
   santoro:
@@ -531,11 +531,11 @@ servers:
     hostgroups: secondary-IPs
   # }}}
   # {{{ gw-conova
-  aagaard:
+  conova-node01:
     address: 217.196.149.227
     parents: gw-conova
     hostgroups: computers, stretch, service, sw-raid
-  acker:
+  conova-node02:
     address: 217.196.149.228
     parents: gw-conova
     hostgroups: computers, stretch, service, sw-raid
@@ -581,11 +581,11 @@ servers:
   csail-node01:
     address: 128.31.0.16
     parents: gw-csail
-    hostgroups: computers, service, dl360, jessie, drbd-hosts
+    hostgroups: computers, service, dl360, stretch, drbd-hosts
   csail-node02:
     address: 128.31.0.46
     parents: gw-csail
-    hostgroups: computers, service, dl360, jessie, drbd-hosts
+    hostgroups: computers, service, dl360, stretch, drbd-hosts
   ganeti-csail:
     address: 128.31.0.49
     parents: gw-bytemark
@@ -644,11 +644,11 @@ servers:
   grnet-node01:
     address: 194.177.211.195
     parents: gw-grnet
-    hostgroups: computers, service, dl380, jessie, drbd-hosts
+    hostgroups: computers, service, dl380, stretch, drbd-hosts
   grnet-node02:
     address: 194.177.211.196
     parents: gw-grnet
-    hostgroups: computers, service, dl380, jessie, drbd-hosts
+    hostgroups: computers, service, dl380, stretch, drbd-hosts
   loghost-grnet-01:
     address: 194.177.211.200
     parents: gw-grnet
@@ -684,7 +684,7 @@ servers:
   static-master-grnet-01:
     address: 194.177.211.209
     parents: ganeti-grnet
-    hostgroups: computers, service, kvmdomains, stretch, systemd-timesyncd
+    hostgroups: computers, service, kvmdomains, stretch, systemd-timesyncd, hassrvfs
   # }}}
   # {{{ gw-isc
   mirror-isc:
@@ -866,7 +866,7 @@ servers:
   byrd:
     address: 140.211.166.200
     parents: gw-osuosl
-    hostgroups: computers, service, dl380, jessie
+    hostgroups: computers, service, dl380, stretch
   beach:
     address: 140.211.166.201
     parents: byrd
@@ -875,7 +875,7 @@ servers:
   pieta:
     address: 140.211.166.195
     parents: gw-osuosl
-    hostgroups: computers, stretch, service
+    hostgroups: computers, stretch, service, manyprocesses
   ppc64el-osuosl-01:
     address: 140.211.166.196
     parents: pieta
@@ -1003,19 +1003,19 @@ servers:
   ubc-enc2bl01:
     address: 209.87.16.1
     parents: ubc-gateway
-    hostgroups: computers, bl460g8, service, jessie, multipath-hosts
+    hostgroups: computers, bl460g8, service, stretch, multipath-hosts, manyprocesses
   ubc-enc2bl02:
     address: 209.87.16.2
     parents: ubc-gateway
-    hostgroups: computers, bl460g8, service, jessie, multipath-hosts
+    hostgroups: computers, bl460g8, service, stretch, multipath-hosts, manyprocesses
   ubc-enc2bl09:
     address: 209.87.16.9
     parents: ubc-gateway
-    hostgroups: computers, bl460g8, service, jessie, multipath-hosts
+    hostgroups: computers, bl460g8, service, stretch, multipath-hosts, manyprocesses
   ubc-enc2bl10:
     address: 209.87.16.10
     parents: ubc-gateway
-    hostgroups: computers, bl460g8, service, jessie, multipath-hosts
+    hostgroups: computers, bl460g8, service, stretch, multipath-hosts, manyprocesses
 
   rachmaninoff:
     address: 209.87.16.20
@@ -1139,7 +1139,7 @@ servers:
   prokofiev:
     address: 177.220.10.140
     parents: gw-unicamp
-    hostgroups: computers, stretch, service
+    hostgroups: computers, stretch, service, manyprocesses
   powerpc-unicamp-01:
     address: 177.220.10.141
     parents: prokofiev
@@ -1349,6 +1349,8 @@ hostgroups:
   broken_https_default_vhost:
     alias: https default vhost does not say 200 OK
     private: 1
+  manyprocesses:
+    alias: hosts with lots and lots of (kernel) processes
 
   no-bacula:
     alias: hosts which are not being backed up with bacula
@@ -1621,11 +1623,11 @@ services:
     name: processes - total
     nrpe: "/usr/lib/nagios/plugins/check_procs 620 700"
     hostgroups: computers
-    excludehosts: prokofiev, pieta, ubc-enc2bl01, ubc-enc2bl02, ubc-enc2bl09, ubc-enc2bl10, fasolo, casulana
+    excludehostgroups: manyprocesses
   -
     name: processes - total
+    hostgroups: manyprocesses
     nrpe: "/usr/lib/nagios/plugins/check_procs 1500 1700"
-    hosts: prokofiev, pieta, ubc-enc2bl01, ubc-enc2bl02, ubc-enc2bl09, ubc-enc2bl10, fasolo, casulana
   -
     name: swap usage - percent
     nrpe: "/usr/lib/nagios/plugins/check_swap -w 20% -c 10%"
@@ -2070,6 +2072,7 @@ services:
     name: unwanted process - rpc.statd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C rpc.statd"
     hostgroups: stretch
+    excludehosts: storace
   -
     name: unwanted process - inetd
     nrpe: "/usr/lib/nagios/plugins/check_procs -w 0:0 -C inetd"