playbooks/nrpe.yml

@@ -24,7 +24,7 @@
         nrpe_command:
           check_disk_all:
             script: check_disk
-            option: -M -u GB -X nfs -X tracefs -X cgroup -X tmpfs -X overlay -X shm -w 15% -c 10% -W 15% -K 10% -A -I '^/run/' -I '^udev$' -I '^/var/lib/kubelet/'
+            option: -M -u GB -X nfs -X tracefs -X cgroup -X tmpfs -X overlay -X shm -w 20% -c 10% -W 20% -K 10% -A -I '^/run/' -I '^udev$' -I '^/var/lib/kubelet/'
           check_load:
             script: check_load
             option: -r -w 0.8,0.8,0.8 -c 1.0,0.9,0.9
@@ -35,9 +35,6 @@
             script: check_users
             option: -w 3 -c 5
         nrpe_command_extra:
-          check_reboot_required:
-            script: /usr/local/bin/monitoring-scripts/check_reboot_required
-            option: 604800
           check_systemd_ansiblepull_service:
             script: /usr/local/bin/monitoring-scripts/check_systemd_unit
             option: ansible-pull.service

playbooks/tasks/web/nagios-hosts.cfg.j2

@@ -108,12 +108,6 @@ define service {
 	check_command		check_nrpe!check_load
 	hostgroup_name		nagios-nrpe
 }
-define service {
-	use			ansible-generic-service
-	service_description	Reboot Required
-	check_command		check_nrpe!check_reboot_required
-	hostgroup_name		nagios-nrpe
-}
 # Systemd unit checks
 define service {
 	use			ansible-generic-service

reboot-home.yml

@@ -1,7 +1,7 @@
 #!/usr/bin/env ansible-playbook
 # vim:ft=ansible:
 ---
-- hosts: tags_pik8s-node,tags_pistorage,tags_pik8s-master
+- hosts: pik8s_nodes,pistorage,pik8s_masters
   serial: 1
   tasks:
     - name: check for reboot-required

reboot-prod.yml

@@ -1,7 +1,7 @@
 #!/usr/bin/env ansible-playbook
 # vim:ft=ansible:
 ---
-- hosts: tags_prod
+- hosts: prod
   serial: 1
   tasks:
     - name: check for reboot-required